System

Abstract

To provide a system for achieving highly accurate task processing while suppressing resource consumption, and for facilitating model customization reflecting knowledge unique to a user.SOLUTION: A system comprising: means for maintaining a base model; means for generating a small derivative model optimized for a specific task; means for providing the small derivative model based on a request from a user; means for receiving user specific knowledge and data and customizing the model; means for storing the customized model; means for distributing the customized model to other users; means for tracking usage of the customized model; means for calculating and providing compensation to the user based on the tracking data; and means for using the model in an offline environment.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The technology of the present disclosure relates to a system. [Background technology]

[0002] Patent document 1 discloses a persona chatbot control method performed by at least one processor, the method including the steps of receiving a user utterance, adding the user utterance to a prompt including an instruction sentence related to a description of the chatbot character, encoding the prompt, and inputting the encoded prompt into a language model to generate a chatbot utterance in response to the user utterance. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-180282 Summary of the Invention [Problem to be solved by the invention]

[0004] Large-scale generative AI models can process a variety of tasks with high accuracy, but they consume large resources and are expensive to dedicate to specific tasks. Furthermore, standard models are difficult to customize to reflect a user's specific knowledge and data, limiting their use in offline environments. Another issue is the lack of a reward system for sharing customized models among users. [Means for solving the problem]

[0005] This invention solves these problems by providing a system that includes a means for maintaining a base model, a means for generating small derivative models optimized for specific tasks, a means for providing a model based on a user's request, a means for customizing the model based on the user's specific knowledge and data, a means for saving the customized model, a means for distributing the customized model to other users, a means for tracking the use of the model, a means for calculating and providing rewards based on the tracking data, and a means for using the model in an offline environment. The system also includes a means for a user to customize the model in a local environment, a means for optimizing the performance of the model, a means for version management of the customized model, and a means for providing an interface for users to share the customized model with other users. This enables high-precision task processing while reducing resource consumption and facilitates model customization that reflects the user's specific knowledge. Furthermore, the system enables use in an offline environment and establishes a reward system between users.

[0006] A "foundation model" is a generative AI model that has been pre-trained using a large dataset and is a general-purpose model that can handle a variety of tasks.

[0007] A "small derivative model" is a scaled-down version of a base model that is optimized for a specific task and is designed to maintain accuracy while reducing resource consumption.

[0008] A "request" is an operation or request that a user makes to a server for a model or function corresponding to a specific task.

[0009] "Customization" refers to a user updating the model's learning using their own specific knowledge and data, adjusting its performance and functionality to suit their purpose.

[0010] An "interface" is a screen, API, or other means by which a user interacts with a server or system, and is the point of contact for data input and operations.

[0011] "Storage" refers to the operations and techniques used to record customized models and data in a system or device so that they can be reused later.

[0012] "Distribution" refers to providing customized models or content to other users or devices.

[0013] "Tracking" means recording and tracking the number and circumstances of use of a model.

[0014] "Reward" refers to compensation such as money or points provided to a user when a model customized by the user is used by another user.

[0015] "Offline environment" refers to the environment in which a system or model is used without connection to the Internet or a server.

[0016] "Version control" refers to the techniques and operations that allow different versions of a customized model to be stored and a specific version to be referenced and restored when needed. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a conceptual diagram showing an example of the configuration of a data processing system according to a first embodiment. [Figure 2] 1 is a conceptual diagram showing an example of main functions of a data processing device and a smart device according to a first embodiment. [Figure 3] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a second embodiment. [Figure 4] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and smart glasses according to a second embodiment. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a third embodiment. [Figure 6]FIG. 11 is a conceptual diagram showing an example of main functions of a data processing device and a headset-type terminal according to a third embodiment. [Figure 7] FIG. 10 is a conceptual diagram showing an example of the configuration of a data processing system according to a fourth embodiment. [Figure 8] FIG. 10 is a conceptual diagram showing an example of main functions of a data processing device and a robot according to a fourth embodiment. [Figure 9] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 10] 1 shows an emotion map onto which multiple emotions are mapped. [Figure 11] FIG. 3 is a sequence diagram showing a processing flow of the data processing system according to the first embodiment. [Figure 12] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 1. [Figure 13] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system according to the second embodiment when an emotion engine is combined. [Figure 14] FIG. 10 is a sequence diagram showing the flow of processing in the data processing system in Application Example 2 when an emotion engine is combined. DETAILED DESCRIPTION OF THE INVENTION

[0018] An example of an embodiment of a system according to the technology of the present disclosure will be described below with reference to the accompanying drawings.

[0019] First, the terms used in the following description will be explained.

[0020] In the following embodiments, a coded processor (hereinafter simply referred to as a "processor") may be a single arithmetic device or a combination of multiple arithmetic devices. Furthermore, a processor may be a single type of arithmetic device or a combination of multiple types of arithmetic devices. Examples of arithmetic devices include a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), a GPGPU (General-Purpose computing on Graphics Processing Units), and an APU (Accelerated Processing Unit).

[0021] In the following embodiments, a coded RAM (Random Access Memory) is a memory in which information is temporarily stored and is used as a working memory by a processor.

[0022] In the following embodiments, the coded storage is one or more non-volatile storage devices that store various programs, various parameters, etc. Examples of non-volatile storage devices include flash memory (SSD (Solid State Drive)), magnetic disks (e.g., hard disks), and magnetic tapes.

[0023] In the following embodiments, a communication I / F (Interface) with a symbol is an interface including a communication processor, an antenna, etc. The communication I / F controls communication between multiple computers. Examples of communication standards applied to the communication I / F include wireless communication standards including 5G (5th Generation Mobile Communication System), Wi-Fi (registered trademark), Bluetooth (registered trademark), etc.

[0024] In the following embodiments, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed connected by "and / or."

[0025] [First embodiment]

[0026] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0027] 1, a data processing system 10 includes a data processing device 12 and a smart device 14. An example of the data processing device 12 is a server.

[0028] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0029] The smart device 14 includes a computer 36, a reception device 38, an output device 40, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The reception device 38, the output device 40, and the camera 42 are also connected to the bus 52.

[0030] The reception device 38 includes a touch panel 38A, a microphone 38B, and the like, and receives user input. The touch panel 38A detects contact with an indicator (for example, a pen or a finger) to receive user input by the touch of the indicator. The microphone 38B detects the user's voice to receive user input by voice. The control unit 46A transmits data indicating the user input received by the touch panel 38A and the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the data indicating the user input.

[0031] The output device 40 includes a display 40A and a speaker 40B, and presents data to the user 20 by outputting the data in a form of expression that the user 20 can perceive (for example, audio and / or text). The display 40A displays visible information such as text and images in accordance with instructions from the processor 46. The speaker 40B outputs audio in accordance with instructions from the processor 46. The camera 42 is a compact digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor.

[0032] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54.

[0033] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0034] 2, in the data processing device 12, a specific process is performed by the processor 28. A specific processing program 56 is stored in the storage 32. The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific process is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0035] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0036] In the smart device 14, the processor 46 performs the reception output process. The storage 50 stores a reception output program 60. The reception output program 60 is used in conjunction with the specific processing program 56 by the data processing system 10. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0037] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0038] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models dedicated to specific tasks and customize them with users' unique knowledge and data. Furthermore, this system can be used offline and has the ability to provide rewards each time a customized model is used by other users.

[0039] Server-side processing

[0040] 1. Creating and serving the model

[0041] The server holds the underlying generative AI model. When a user requests a smaller derivative model optimized for a specific task (e.g., text generation in a specific domain), the server generates the derivative model according to the request and provides it to the user.

[0042] 2. Managing Customizations

[0043] The server provides a customization interface. Users upload their own specific knowledge and data, and the server receives the data and retrains or fine-tunes the specified derived model to suit the user's needs. The customized model is saved as a new version.

[0044] 3. Model Distribution and Reward Management

[0045] The server provides a function for distributing customized models to other users. Each time other users use a customized model, the server tracks the usage and calculates rewards based on the tracking data. Rewards are paid to the user who provided the customized model.

[0046] Terminal side processing

[0047] 1. Customizing the model

[0048] The user's device receives the derived model provided by the server and customizes it using local data (e.g., reviews of a particular product). The customization process is performed on the device, and the customized model is saved locally.

[0049] 2. Use in offline environments

[0050] The device has the ability to run customized models offline, for example, a product review generation model can run offline and perform tasks even when there is no internet connection.

[0051] User processing

[0052] 1. Requests and Customizations

[0053] A user requests a derived model for a specific task from the server, then customizes the model by uploading specific knowledge and data using the provided interface, and the customized model is stored locally on the user's computer.

[0054] 2. Use of Models and Receipt of Payments

[0055] Users can share their customized models with other users, and the server tracks usage and provides rewards each time other users use the model.

[0056] Specific examples

[0057] 1. Server

[0058] The server holds a basic model for "medical report generation." When User A requests a report generation model specialized for a specific medical condition, the server generates a derived model corresponding to the request and provides it to User A.

[0059] 2. Terminal

[0060] User A customizes the derived model on the device using local data acquired from a specific hospital. The customized model is saved on the device and can be used offline.

[0061] 3. Users

[0062] User A distributes a customized "condition-specific report generation model" online. When User B uses the model to generate a condition report, the server tracks its usage and provides a reward to User A.

[0063] In this way, a system is realized in which servers, terminals, and users can cooperate to efficiently and flexibly utilize generative AI models, creating, distributing, and using models optimized for specific tasks.

[0064] The processing flow will be explained below.

[0065] Server-side processing

[0066] Step 1:

[0067] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[0068] Step 2:

[0069] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[0070] Step 3:

[0071] The server provides the generated small derivative model to the user.

[0072] Step 4:

[0073] The server provides a customization interface where users can upload their own knowledge and data.

[0074] Step 5:

[0075] The server uses the received data to retrain or fine-tune the derived model to suit the user's needs.

[0076] Step 6:

[0077] The server saves the customized model as a new version.

[0078] Step 7:

[0079] The server configures the customized model for distribution to other users.

[0080] Step 8:

[0081] Each time another user uses the customized model, the server tracks that usage.

[0082] Step 9:

[0083] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[0084] Terminal side processing

[0085] Step 1:

[0086] The terminal receives the derived model provided by the server.

[0087] Step 2:

[0088] The user's device loads the specific data stored locally.

[0089] Step 3:

[0090] The device uses the provided model to retrain or fine-tune based on local data.

[0091] Step 4:

[0092] The device stores the customized model locally.

[0093] Step 5:

[0094] The device runs the customized model in an offline environment to perform tasks.

[0095] User processing

[0096] Step 1:

[0097] A user requests a derived model corresponding to a particular task from the server.

[0098] Step 2:

[0099] Users upload their own knowledge and data through a provided interface.

[0100] Step 3:

[0101] The user receives the customized model from the server and uses it.

[0102] Step 4:

[0103] Users can set up sharing of their customized models with other users.

[0104] Step 5:

[0105] See and receive rewards when other users use your customized model.

[0106] Specific examples

[0107] Server-side example

[0108] Step 1:

[0109] User A sends a request for a "medical condition report generation model."

[0110] Step 2:

[0111] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[0112] Step 3:

[0113] The server provides the generated "medical condition report generation model" to User A.

[0114] Terminal side example

[0115] Step 1:

[0116] User A's terminal receives the "medical condition report generation model" provided by the server.

[0117] Step 2:

[0118] User A's terminal reads the medical condition data acquired from a specific hospital.

[0119] Step 3:

[0120] The device retrains the provided model using the "medical condition data" and generates a customized "medical condition report generation model."

[0121] Step 4:

[0122] The customized model is saved on User A's device.

[0123] User-side example

[0124] Step 1:

[0125] User A sets up the customized model to be publicly available to other users.

[0126] Step 2:

[0127] User B uses the model to generate a medical report.

[0128] Step 3:

[0129] The server tracks User B's usage and provides a reward to User A.

[0130] Example 1

[0131] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0132] In order to quickly and individually respond to today's diverse user needs, it is desirable to use generative AI models optimized for specific tasks. However, existing systems make it difficult for users to customize models using their own knowledge and data, and it is not easy to use them in offline environments, share customized models, or calculate rewards. This hinders the efficient and flexible use of generative AI models.

[0133] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0134] In this invention, the server includes: means for storing a base model; means for generating a compact derivative model optimized for a specific task; means for providing the compact derivative model based on a user request; means for receiving user-specific knowledge and data and customizing the model; means for saving the customized model; means for distributing the customized model to other users; means for tracking use of the customized model; means for calculating and providing a reward to the user based on the tracking data; means for using the model in an offline environment; means for customizing the model provided by the server in a local environment; means for a user to upload data through an interface and retrain the model based on the data; means for tracking usage of the generated customized model each time it is used by other users and calculating a reward; and means for providing the reward calculated based on the usage as credits to the user's account. This enables users to easily and flexibly customize, use, and share generative AI models tailored to their specific needs.

[0135] A "foundation model" is a large-scale generative AI model that can handle a variety of tasks.

[0136] A "small derivative model" is a small, specialized generative AI model that optimizes a base model for a specific task.

[0137] A "request" is a user's request for the generation of a derived model corresponding to a particular task.

[0138] "Customization" is the process of adapting an existing model using a user's specific knowledge and data.

[0139] "Saving" is the act of permanently recording the generated model data so that it can be accessed later.

[0140] "Distribution" is the act of providing a customized model to other users.

[0141] "Tracking" means monitoring and recording the usage and performance of a model.

[0142] "Reward" is the compensation received by a user who provides a customized model.

[0143] An "offline environment" is a situation where there is no internet connection.

[0144] A "local environment" is a system environment that runs on a user's terminal.

[0145] An "interface" is the means or tool by which a user interacts with a system.

[0146] "Retraining" is the process of updating an existing model with new data.

[0147] An "account" is an individual identification information that is used when a user is registered in the system and information such as rewards is managed.

[0148] "Credit" is a unit or form of reward used within the system.

[0149] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models optimized for specific tasks, which can then be customized with the user's specific knowledge and data. Additionally, this system can be used offline and has the ability to provide rewards each time the customized model is used by other users.

[0150] The server stores the base model, and when a user requests a small derived model optimized for a specific task (e.g., generating medical reports or product reviews), the server generates a derived model adapted to the specific task from the base model and provides it to the user. The server performs this procedure using a deep learning framework such as TensorFlow or PyTorch.

[0151] After receiving the derived model from the server, the user customizes it in their local environment (on their device). The user's device receives the derived model provided by the server and retrains the model using local data (specific medical records, product reviews, etc.). The retraining process is also performed using TensorFlow, PyTorch, etc. The customized model is saved in the user's local environment, allowing it to be used in offline environments.

[0152] The server receives the specific knowledge and data uploaded by the user and provides an interface for customization. Once the user customizes the model, the server saves the new version of the model and provides settings for distribution to other users. The server also tracks the usage of the customized model each time it is used by other users, calculates rewards based on the number and frequency of use, and adds them as credits to the user's account.

[0153] As a specific example, if a server holds a basic model for "medical report generation," when User A requests a report generation model specialized for a specific medical condition, the server generates a derived model according to the request and provides it to User A. User A customizes the derived model on his terminal using medical data obtained from a specific hospital and uses the customized model in offline environments. Furthermore, when User A sets up online distribution of the customized "medical condition-specific report generation model" and User B uses that model to generate a medical condition report, the server tracks its usage and provides User A with a reward.

[0154] An example prompt might be, "I would like to request an AI model to generate medical reports specific to the symptoms of rheumatoid arthritis patients. Furthermore, I would like to customize the model using specific hospital data."

[0155] In this way, by coordinating the functions of the server, terminal, and user, it becomes possible to efficiently and flexibly generate, use, and share generative AI models optimized for specific tasks.

[0156] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0157] Step 1:

[0158] The server receives requests from users for derivative models corresponding to particular tasks.

[0159] Input: User request information (e.g., "Medical report generation model specialized for rheumatoid arthritis patients")

[0160] Specific operation: The user sends a request for a generative AI model to the server. The expected prompt is, "I would like to request an AI model that generates medical reports specific to the symptoms of rheumatoid arthritis patients."

[0161] Step 2:

[0162] The server uses the underlying generative AI model to generate smaller derivative models optimized for specific tasks.

[0163] Input: Base model stored on the server, user request information

[0164] Output: Smaller variants optimized for specific tasks

[0165] Specific operation: The server uses deep learning frameworks such as TensorFlow and PyTorch to generate derived models from the base model according to the request.

[0166] Step 3:

[0167] The server provides the generated derived model to the user.

[0168] Input: Generated small derivative model

[0169] Output: Derived models provided to users (via download link or API)

[0170] Specific operation: The server provides the derived model to the user by sending a download link via email or by providing an API endpoint.

[0171] Step 4:

[0172] The user's terminal receives the derived model provided by the server and customizes it with local data.

[0173] Input: Provided derived model, user's local data (e.g. medical records)

[0174] Output:Customized model

[0175] Specific operation: The user's device acquires local data (e.g., medical data obtained from a specific hospital) and retrains the model using TensorFlow or PyTorch.

[0176] Step 5:

[0177] The server manages the customization process using data provided by the user.

[0178] Input: User-uploaded specific data, customization interface

[0179] Output: Retrained customization model, notification of customization completion

[0180] Specific operation: The server receives the data uploaded by the user, performs the re-training process, and then saves the new version of the customized model.

[0181] Step 6:

[0182] The server provides the configuration to distribute customized models to other users.

[0183] Input: Customized model, distribution settings request

[0184] Output: Distribution settings to other users

[0185] Specific operation: The server receives a distribution configuration request from the user, performs the configuration process, and generates a public link and API endpoint.

[0186] Step 7:

[0187] The server tracks usage and calculates rewards each time the customized model is used by other users.

[0188] Input: Model usage tracking data

[0189] Output: Usage tracking record, calculated rewards

[0190] How it works: The server tracks usage of API calls, etc., and calculates rewards based on the number and frequency of usage.

[0191] Step 8:

[0192] The server provides the calculated reward as a credit to the user's account.

[0193] Input: Calculated reward

[0194] Output: Credits added to the user's account

[0195] Specific operation: The server adds the calculated reward to the user's account as credits, which the user can check on the dashboard etc.

[0196] Step 9:

[0197] The user terminal uses the customized model in an offline environment.

[0198] Input:Customized Model

[0199] Output: offline generated tasks (e.g. offline generated medical reports)

[0200] Specific operation: Users can use customized models, even without an internet connection, to generate medical reports, for example.

[0201] (Application example 1)

[0202] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0203] For autonomous vehicles, route optimization and emergency response to real-time road conditions and emergency situations are important. However, it is difficult to easily utilize the customized AI models required to meet these requirements in current systems. There is also a need for models that can be used in offline environments. Furthermore, there is a lack of functionality to track the usage of various customized models and provide appropriate rewards to users.

[0204] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0205] In this invention, the server includes means for storing a base model, means for generating a compact derivative model optimized for a specific task, means for providing the compact derivative model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for using the model in an offline environment, and means for generating and customizing a derivative model optimized for a specific task using the generative AI model to support route optimization and emergency response in autonomous vehicles. This improves the accuracy of route optimization and emergency response in autonomous vehicles and significantly improves user convenience.

[0206] A "foundation model" is a fundamental AI model that plays a key role in the system and serves as the basis for derivative models optimized for specific tasks.

[0207] A "small derivative model" is a lightweight, efficient AI model generated from a base model and optimized for a specific task.

[0208] "User" refers to an individual or organization that utilizes the system to request and customize models optimized for specific tasks.

[0209] A "request" refers to an action in which a user asks the system to generate a derived model specialized for a specific task.

[0210] "Customization" is the process of adjusting a model using a user's specific knowledge and data to optimize it for their specific needs and requirements.

[0211] An "offline environment" refers to a state in which an Internet connection is not established, and the system operates independently of a network.

[0212] "Route optimization" refers to the calculations and processes that an autonomous vehicle performs to select the optimal route to its destination.

[0213] "Emergency response" refers to the process of providing appropriate handling and countermeasures for emergencies such as accidents and breakdowns.

[0214] A "generative AI model" refers to an artificial intelligence model that generates sentences, images, etc. based on large amounts of data.

[0215] "Tracking" refers to the act of tracking the use of a Customized Model and collecting Usage Data.

[0216] "Reward" refers to the compensation paid to the provider when the customized model is used by other users.

[0217] An "autonomous vehicle" refers to a vehicle that drives autonomously using sensors and artificial intelligence technology.

[0218] The present invention relates to a system for supporting route optimization and emergency response in an autonomous driving vehicle, and specific embodiments thereof will be described below.

[0219] Server-side processing

[0220] The server maintains the underlying model and generates smaller derivative models optimized for specific tasks. When a user submits data and a request, the server processes it using the following methods:

[0221] 1. Creating and serving the model

[0222] Using a base generative AI model (e.g., GPT-3), a derivative model is generated based on the user's request and provided to the user.

[0223] 2. Managing Customizations

[0224] It provides a customization interface and receives user-specific knowledge and data to retrain or fine-tune derived models, while also referencing real-time data from Google Maps API, Weather API, etc.

[0225] 3. Model Distribution and Reward Management

[0226] The usage of the customized model is tracked and rewards are calculated based on the usage data, and rewards are provided to the user who provided the customized model.

[0227] Terminal side processing

[0228] Users also have the ability to customize and save models locally on their devices that are optimized for specific tasks.

[0229] 1. Customizing the model

[0230] The user's device receives the derived model provided by the server and customizes it using local data (e.g., traffic information for a specific area). This customization process is performed on the device to optimize the performance of the generated AI model.

[0231] 2. Use in offline environments

[0232] The customized model can also work offline, allowing it to perform tasks even when there is no internet connection.

[0233] User processing

[0234] Users use the system to request and customize derivative models for specific tasks.

[0235] 1. Requests and Customizations

[0236] Users submit requests to the server and upload specific data to customize the model, generating prompts and entering the required data.

[0237] 2. Use of Models and Receipt of Payments

[0238] Users can share their customized models with other users and receive rewards each time they use the model.

[0239] Specific examples

[0240] When a user travels from Tokyo to Yokohama, the following process takes place:

[0241] 1. Example of route optimization

[0242] The user enters their current location "Tokyo" and their destination "Yokohama."

[0243] Provides real-time data on "heavy traffic" and "sunny" weather.

[0244] Example prompt: "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Generate the optimal route."

[0245] The generated optimal route is returned.

[0246] 2. Specific examples of emergency response

[0247] A user reports an emergency condition "accident" and specifies the location "Highway 246."

[0248] Example prompt: "Emergency situation: Accident. Location: Highway 246. Generate the appropriate response."

[0249] The generated contingency plan is provided.

[0250] In this way, the system of the present invention can improve the accuracy of route optimization and emergency response in autonomous vehicles, significantly improving user safety and convenience.

[0251] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0252] Step 1:

[0253] A user opens an autonomous vehicle application and sends a request for route optimization or emergency response to the server. The inputs include current location, destination, and real-time data (e.g., traffic conditions, weather information). The server receives this request and generates a prompt for the underlying generative AI model (e.g., GPT-3).

[0254] Step 2:

[0255] The server queries the base generative AI model using the generated prompt. An example of a generated prompt is "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Please generate the optimal route." The prompt is given as input, and the optimal route is generated as output.

[0256] Step 3:

[0257] The server provides the generated optimal route to the user, which then returns the route information to the user's device. The user can review this route information and customize the model locally as needed. The generated route information is used as input and displayed on the user's device as output.

[0258] Step 4:

[0259] The user takes the route generated on the device and customizes it based on their specific circumstances, for example, modifying the route to take into account specific traffic patterns or personal driving preferences. User-specific data is used as input, and a customization model is generated as output.

[0260] Step 5:

[0261] The device stores the customized model and allows it to run offline if necessary. The input is the customized model and the output is the saved model. This allows the model to run even without an internet connection.

[0262] Step 6:

[0263] When a user encounters a situation requiring emergency response, the device sends an emergency request to the server. The input includes the type of emergency situation (e.g., accident) and location information. The server receives this request and generates a prompt.

[0264] Step 7:

[0265] The server generates emergency response measures based on the generated prompt text. An example of a prompt text is "Emergency situation: Accident Location: Highway 246. Please generate appropriate response measures." The prompt text is given as input, and an emergency response measure is generated as output.

[0266] Step 8:

[0267] The server returns the generated contingency plan to the user and displays it on the user's device. The user can take appropriate action based on this information. The generated contingency plan is used as input and displayed on the user's device as output.

[0268] Step 9:

[0269] The server tracks all customized models and their usage, and calculates rewards, using the tracking data as input and generating reward information as output, which is provided to the user.

[0270] Furthermore, an emotion engine that estimates the user's emotion may be combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0271] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks, can be customized with users' specific knowledge and data, and is combined with an emotion engine that recognizes users' emotions. This system can also be used in offline environments and has the function of providing rewards each time the customized model is used by other users.

[0272] Server-side processing

[0273] 1. Creating and serving the model

[0274] The server maintains the base model and receives requests from users for derived models for specific tasks (e.g., text generation in a specific field). Based on the request, the server generates a small derived model from the base model that is optimized for the specific task and provides it to the user.

[0275] 2. Customization management and emotion engine integration

[0276] The server provides a customization interface where users can upload their specific knowledge and data. It then uses an emotion engine to recognize the user's emotions and optimizes the customization model based on that data. The received data is then used to retrain and fine-tune the model to suit the user.

[0277] 3. Model Distribution and Reward Management

[0278] The server saves the customized model as a new version and configures it for distribution to other users. Each time another user uses the customized model, it tracks its usage, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[0279] Terminal side processing

[0280] 1. Customizing the model and using emotion data

[0281] The user's device receives the derived model provided by the server, reads the specific data (e.g., specific product review information) stored locally, and uses the emotion engine to obtain the user's emotion data and customize the model based on that data. The customized model is stored locally.

[0282] 2. Use in offline environments

[0283] The device has the ability to run customized models offline, allowing tasks to be performed even when there is no internet connection.

[0284] User processing

[0285] 1. Requests and Customizations

[0286] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided interface, and customize the models using emotion data obtained from the emotion engine.

[0287] 2. Use of Models and Receipt of Payments

[0288] Users can share their customized models with other users and receive rewards each time they use the model.

[0289] Specific examples

[0290] Server-side example

[0291] 1. Receiving a request and providing a model

[0292] When user A requests a "medical condition report generation model," the server receives the request, generates a derived model specialized for "medical condition report generation" from the base model, and provides it to user A.

[0293] 2. Customization management and emotion engine integration

[0294] User A uploads data obtained from a specific hospital, and the server uses an emotion engine to analyze User A's emotion data. Based on the analysis results, the derived model is retrained to generate a customized "condition report generation model."

[0295] Terminal side example

[0296] 1. Customizing the model

[0297] User A's device receives the "medical condition report generation model" provided by the server, reads the locally stored medical condition data, and uses the emotion engine to obtain User A's emotional data and optimize the model based on that data.

[0298] 2. Use in offline environments

[0299] User A's terminal executes the customized "medical condition report generation model" in an offline environment to generate a medical condition report.

[0300] User-side example

[0301] 1. Request and Data Upload

[0302] User A requests a "condition report generation model" from the server and uploads data from a specific hospital. Furthermore, the model is customized using data from the emotion engine.

[0303] 2. Use of Models and Receipt of Payments

[0304] User A sets up a customized "medical condition report generation model" to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[0305] In this way, a system is realized in which servers, devices, users, and emotion engines cooperate to efficiently and flexibly utilize generative AI models, and create, distribute, and use models optimized for specific tasks.

[0306] The processing flow will be explained below.

[0307] Server-side processing

[0308] Step 1:

[0309] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[0310] Step 2:

[0311] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[0312] Step 3:

[0313] The server provides the generated small derivative model to the user.

[0314] Step 4:

[0315] The server provides a customization interface where users can upload their own knowledge and data.

[0316] Step 5:

[0317] The server uses an emotion engine to collect user emotion data.

[0318] Step 6:

[0319] The server retrains or fine-tunes the derived model based on the user's specific data and emotional data.

[0320] Step 7:

[0321] The server saves the customized model as a new version.

[0322] Step 8:

[0323] The server configures the customized model for distribution to other users.

[0324] Step 9:

[0325] Each time another user uses the customized model, the server tracks that usage.

[0326] Step 10:

[0327] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[0328] Terminal side processing

[0329] Step 1:

[0330] The terminal receives the derived model provided by the server.

[0331] Step 2:

[0332] The user's device loads the specific data stored locally.

[0333] Step 3:

[0334] The terminal uses an emotion engine to obtain emotion data of the user.

[0335] Step 4:

[0336] The device uses the provided model to retrain or fine-tune based on local data and emotion data.

[0337] Step 5:

[0338] The device stores the customized model locally.

[0339] Step 6:

[0340] The device runs the customized model in an offline environment to perform tasks.

[0341] User processing

[0342] Step 1:

[0343] A user requests a derived model corresponding to a particular task from the server.

[0344] Step 2:

[0345] Users upload their own knowledge and data through a provided interface.

[0346] Step 3:

[0347] The user provides emotion data through the emotion engine.

[0348] Step 4:

[0349] The user receives the customized model from the server and uses it.

[0350] Step 5:

[0351] Users can set up sharing of their customized models with other users.

[0352] Step 6:

[0353] See and receive rewards when other users use your customized model.

[0354] Specific examples

[0355] Server-side example

[0356] Step 1:

[0357] User A sends a request for a "medical condition report generation model."

[0358] Step 2:

[0359] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[0360] Step 3:

[0361] The server provides the generated "medical condition report generation model" to User A.

[0362] Step 4:

[0363] The server provides a customization interface where User A uploads specific hospital data.

[0364] Step 5:

[0365] The server uses an emotion engine to collect emotion data of user A.

[0366] Step 6:

[0367] The server retrains the derived model based on the uploaded data and emotion data of user A.

[0368] Step 7:

[0369] The server stores the customized model and serves it as a new version.

[0370] Terminal side example

[0371] Step 1:

[0372] User A's terminal receives the "medical condition report generation model" provided by the server.

[0373] Step 2:

[0374] User A's terminal reads the medical condition data acquired from a specific hospital.

[0375] Step 3:

[0376] User A's device uses an emotion engine to acquire User A's emotion data.

[0377] Step 4:

[0378] The device retrains using medical condition data and emotion data to generate a customized model.

[0379] Step 5:

[0380] The customized model is saved on User A's device.

[0381] Step 6:

[0382] User A's terminal generates a medical condition report even in an offline environment.

[0383] User-side example

[0384] Step 1:

[0385] User A requests the "medical condition report generation model" from the server.

[0386] Step 2:

[0387] User A uploads hospital data through the server interface.

[0388] Step 3:

[0389] User A provides his / her own emotion data through the emotion engine.

[0390] Step 4:

[0391] User A receives the customized model from the server and uses it.

[0392] Step 5:

[0393] User A sets up the customized model to be publicly available to other users.

[0394] Step 6:

[0395] User B uses the model to generate a medical condition report.

[0396] Step 7:

[0397] The server tracks user B's usage and provides a reward to user A.

[0398] Example 2

[0399] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0400] Conventional generative AI models are based on large datasets, making them less versatile for specific tasks and difficult to customize to meet individual needs. They also lack the ability to reflect user emotions in real time, making personalization based on user emotions difficult. Furthermore, they are limited to offline use, requiring a constant internet connection.

[0401] The specific processing by the specific processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a compact derived model optimized for a specific task, means for providing the compact derived model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for acquiring user emotion data using an emotion engine and using it for customization, and means for using the model in an offline environment. This enables the generation of a model tailored to individual needs for a specific task, personalization based on the user's emotions, and use in an offline environment.

[0402] A "foundation model" refers to a large-scale generative AI model, a general-purpose model that can handle a variety of tasks.

[0403] "Small derivative models" refer to lightweight models that are optimized for specific tasks from the base model.

[0404] A "user request" refers to a request from a user to generate a derived model corresponding to a particular task.

[0405] "Customization interface" refers to the operation screen and input means that allow users to upload their own specific knowledge and data and customize the model.

[0406] An "emotion engine" is a software module that acquires and analyzes user emotional data and optimizes the model based on that data.

[0407] "Offline environment" refers to a state where there is no internet connection, and refers to the ability to run models in such an environment.

[0408] "Tracking Data" refers to data (such as the number of uses and duration) when a customized model is used by other users.

[0409] "Remuneration" refers to the consideration paid to a user who provides a customized model in accordance with the use of the model.

[0410] "Local environment" refers to the user's device and its internal storage, an environment where data is stored without going through the Internet.

[0411] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks and can be customized with the user's specific knowledge and data. It also incorporates an emotion engine that recognizes the user's emotions, making it usable in offline environments. It also has a function that provides rewards each time the customized model is used by other users.

[0412] Server-side processing

[0413] The server first stores a large-scale base generative AI model (e.g., GPT-3). This base model is a general-purpose model that can handle a variety of tasks. When a user requests a derived model for a specific task (e.g., generating a medical diagnosis report), the server generates a small, task-optimized derived model from the base model and provides it to the user.

[0414] For example, when user A requests a "medical condition report generation model," the server generates a derived model specialized for "medical condition report generation" from the base model and provides it to user A.

[0415] The server then provides a customization interface, allowing users to upload their own specific data (e.g., data obtained from hospitals or expertise), and utilizes an emotion engine to analyze the user's emotion data and optimize the model based on this data, resulting in a more personalized model.

[0416] The customized model is saved in the database and managed as a new version. The server tracks the usage of this model by other users, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[0417] Terminal side processing

[0418] The device downloads the derived model provided by the server and stores it locally. It also reads specific data (e.g., specific product review information) stored locally on the device. Furthermore, the device uses an emotion engine to obtain user emotion data through input devices such as a camera or microphone, and customizes the derived model based on that data.

[0419] The device also has the ability to run customized models offline, allowing it to complete tasks even when there is no internet connection.

[0420] For example, User A's device locally stores a "medical condition report generation model" and optimizes it by reading local medical condition data and emotion data. This customized model can be used to generate medical condition reports even in offline environments.

[0421] User processing

[0422] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided customization interface, and further customize the models using emotion data obtained from the emotion engine to optimize their performance.

[0423] Users can share their customized models with other users and receive rewards each time they use the model. The server calculates and provides rewards to users based on tracking data.

[0424] Examples of concrete examples and prompts

[0425] For example, User A requests a "medical condition report generation model" and uploads data from a specific hospital to the server. Furthermore, he customizes the model using data from the emotion engine and sets it up to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[0426] Example prompt sentence:

[0427] Generate a medical report: "Generate a medical report based on the following data: [patient information]"

[0428] Product Review Analysis: "Analyze the following review and extract positive and negative comments. Review: [Product Review Text]"

[0429] As described above, this invention provides a flexible system that can be used in offline environments by using a base generation AI model to generate customized derivative models specialized for specific tasks and optimizing them using user emotional data and specific knowledge.

[0430] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0431] Step 1: A user requests a derived model for a specific task.

[0432] Input: A task request sent by a user through a web interface or a dedicated application. A specific example is a medical condition report generation model.

[0433] Processing: The request information is sent to the server, which receives the request and starts processing according to the request contents.

[0434] Output: The server sends a confirmation message back to the user indicating that it has received the request.

[0435] Step 2: The server generates a derived model from the base model.

[0436] Input: A request for a specific task received from a user.

[0437] Processing: The server uses a large-scale generative AI model (e.g., GPT-3) as a base to generate a smaller derivative model optimized for a specific task based on the request. Here, the program sets parameters related to the specific task and calls the generative AI model to create the derivative model.

[0438] Output: The generated derived model file.

[0439] Step 3: The server provides the derived model to the user.

[0440] Input: The generated derived model file.

[0441] Processing: The server generates a download link to provide the model to the user and sends the link information to the user. The model file is also temporarily saved on the server.

[0442] Output: A notification message with a download link.

[0443] Step 4: User uploads data using a customization interface

[0444] Input: User specific data (e.g., hospital report data and expertise) and emotional data.

[0445] Processing: The user uploads their own data and emotion data to the server using the customization interface. The emotion data is retrieved by the emotion engine.

[0446] Output: Files of characteristic data and emotion data sent to the server.

[0447] Step 5: The server analyzes the emotion data and retrains the model

[0448] Input: User-uploaded characteristic and emotion data.

[0449] Processing: The emotion engine analyzes the uploaded emotion data and retrains and customizes the derived model based on the analysis results. Retraining is performed using a dedicated algorithm based on the base model and uploaded data.

[0450] Output: Customized derivative model.

[0451] Step 6: The server saves the customized model and prepares it for deployment.

[0452] Input: Customized derivative model.

[0453] Processing: Save the customized model as a new version in the database and configure it for distribution to other users.

[0454] Output: Saved model data and distribution settings information.

[0455] Step 7: Track when others use your model

[0456] Input: Use models customized by other users.

[0457] Processing: The server tracks the usage of the model (number of uses, duration of use, etc.). This process is achieved by recording usage logs in a database.

[0458] Output: Tracking data (usage log).

[0459] Step 8: The server calculates and provides the reward

[0460] Input: Tracking data (usage logs).

[0461] Processing: The server calculates the reward for the provider of the customized model based on the tracking data, and the calculated reward is deposited into the user's account.

[0462] Output: A record of the reward payment and a payment confirmation message.

[0463] Step 9: The device downloads the derivative model and saves it locally.

[0464] Input: Download link.

[0465] Processing: The user's device downloads the derived model using the download link provided by the server and stores it in local storage.

[0466] Output: Locally saved model file.

[0467] Step 10: The device loads local data and acquires emotion data

[0468] Input: Locally stored characteristic data and real-time emotion data.

[0469] Processing: The user's device reads the locally stored specific data and uses the emotion engine to obtain the user's real-time emotion data. Emotion data is collected using input devices such as cameras and microphones and analyzed.

[0470] Output: Parsed sentiment data and characteristic data.

[0471] Step 11: The device customizes the model based on the acquired data

[0472] Input: Feature data and emotion data.

[0473] Processing: The device uses this data to customize a derived model, adjusting the model parameters in the local environment to optimize it for the user's individual needs.

[0474] Output:Customized model.

[0475] Step 12: Run the customized model in an offline environment

[0476] Input:Customized model.

[0477] Processing: The user's device can execute customized models to accomplish specific tasks even in offline environments without an internet connection.

[0478] Output: The result of the task execution (e.g., the generated medical condition report).

[0479] (Application example 2)

[0480] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart device 14 will be referred to as a "terminal."

[0481] Conventional generative AI models have difficulty creating small models specifically tailored to specific tasks, and the results obtained by using general-purpose models are often inaccurate or insufficient. Furthermore, customization using user-specific knowledge and data is limited, and because emotional data is not taken into account, incentives and reward systems based on user emotions are not fully utilized. Furthermore, because they cannot be used in offline environments, flexible operation that does not rely on the Internet is required. To solve these issues, a more effective and flexibly customizable generative AI model system is needed.

[0482] The specific processing by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a small derived model optimized for a specific task, means for receiving user-specific knowledge and data and customizing the model, means for acquiring user emotional data and using it to customize the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing rewards to the user based on the tracking data, means for using the model in an offline environment, and means for making recommendations as a smartphone app based on the user's purchase history and emotional data. This makes it possible to generate a derived model specialized for a specific task by utilizing the user's specific knowledge and emotional data, and to operate flexibly in an offline environment. Furthermore, providing rewards to users can increase their motivation.

[0483] A "foundation model" is a large-scale model that is central to a generative AI system and serves as the basis for generating smaller derivative models that are specialized for specific tasks.

[0484] A "small derivative model" is a small generative AI model that is optimized from the base model to specialize in a specific task.

[0485] "Specific knowledge and data" refers to information and data sets that are unique to each individual user and are used to customize the model.

[0486] "Emotion data" is data that represents the user's emotional state and is acquired through the emotion engine.

[0487] "Customization" is the process of optimizing a generative AI model for a specific user or task based on specific knowledge, data, and emotional data.

[0488] "Saving" refers to storing a generated or customized model in digital storage for later use.

[0489] "Distribution" refers to providing a customized model to other users, and is done over a communications network.

[0490] "Usage Tracking" is the process of recording and monitoring how your customized model is being used.

[0491] A "reward" is an incentive paid to the provider of a customized model when the customized model is used by another user.

[0492] "Offline environment" refers to an environment without an internet connection, and the model is required to function in this environment.

[0493] "Recommendation" refers to recommending specific products or services based on a user's purchasing history and emotional data.

[0494] A "smartphone app" is application software that runs on a smartphone and performs specific functions.

[0495] This invention is a system based on a large-scale generative AI model that generates smaller derivative models specialized for specific tasks and customizes them with the user's specific knowledge and data. Furthermore, by combining it with an emotion engine that recognizes the user's emotions, it provides a personalized user experience. This system can also be used offline and has the ability to provide rewards each time the customized model is used by other users.

[0496] Overview

[0497] Server-side processing

[0498] 1. Generate and serve the model:

[0499] The server maintains the base model and generates smaller derivative models optimized for specific tasks in response to user requests. When a user makes a specific request, the server creates a corresponding derivative model based on that request and provides it to the user.

[0500] 2. Customization management and emotion engine integration:

[0501] The server provides a customization interface, allowing users to upload their own knowledge and data, and then uses an emotion engine to analyze the user's emotion data and optimize the model based on that data. Through retraining and fine-tuning, a personalized model is generated.

[0502] 3. Model Distribution and Reward Management:

[0503] The server stores and manages customized models, makes them available for distribution to other users, monitors model usage based on tracking data, and calculates and pays rewards to model providers.

[0504] Terminal side processing

[0505] 1. Customizing the model and using emotion data:

[0506] The user's device receives the derived model provided by the server, reads the specific data stored locally (e.g., purchase history), and uses the emotion engine to obtain the user's emotion data and use it to optimize the model.

[0507] 2. Offline use:

[0508] The device allows customized models to run offline, allowing tasks to be performed even when there is no internet connection.

[0509] User processing

[0510] 1. Requests and Customization:

[0511] Users request derived models for specific tasks from the server and upload their specific knowledge and data through the provided interface, and also utilize emotional data obtained from the emotion engine to customize the models.

[0512] 2. Use of Models and Receipt of Payments:

[0513] Users can share their customized models with other users and receive rewards each time their models are used.

[0514] Processing Description

[0515] server

[0516] The server uses the following hardware and software:

[0517] Hardware: High-performance server machine

[0518] Software: Emotion engine, AI model generation library, database management system

[0519] The server maintains the base model and generates derivative models specialized for specific tasks, provides a customization interface, optimizes the model based on user-specific and emotional data, and tracks the use of customized models and calculates rewards.

[0520] Terminal

[0521] The device uses the following hardware and software:

[0522] Hardware: Smartphone

[0523] Software: Emotion engine library, model optimization module

[0524] The device receives the derived model provided by the server and customizes it based on local data and emotion data, allowing it to perform tasks even in offline environments.

[0525] User

[0526] Users can request derived models for specific tasks, upload their specific knowledge and data to the server, and receive optimal recommendations using the customized models, leveraging sentiment data.

[0527] Specific examples

[0528] As an example of usage, consider a situation in which a user receives a recommendation for a specific product through a smartphone app.

[0529] Users can receive personalized recommendations based on their purchasing history and emotional data.

[0530] An example of a prompt sentence is, "Recommend the best products based on the user's purchasing history and current emotional state."

[0531] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[0532] Step 1:

[0533] The server holds the foundation model and receives requests from users. The requests include a specific task the user needs (e.g., recommendations based on purchase history). The input data is the user's request, and the output data is a task instruction for generating a derived model. Based on the task instruction, the server prepares to generate a derived model from the foundation model.

[0534] Step 2:

[0535] The server generates a small derived model optimized for a specific task. The input data here are the base model and the user's request, and the output data is a derived model suitable for the specific task. The derived model is generated from the base model and provided to the user.

[0536] Step 3:

[0537] The terminal receives the derived model provided from the server. The input data is the provided derived model, and the output data is the derived model stored in the terminal. In this step, the terminal performs an operation of locally storing the derived model.

[0538] Step 4:

[0539] The user uploads specific knowledge and data (e.g., purchase history data) to the terminal. The input data is the user's specific data, and the output data is customized data stored locally. The terminal receives and stores the data from the user.

[0540] Step 5:

[0541] The terminal uses an emotion engine to obtain the user's current emotion data. The input data is the user's real-time emotional state, and the output data is emotion data. The emotion engine analyzes the user's emotion and generates the emotion data.

[0542] Step 6:

[0543] The terminal customizes the derived model based on the user's specific data and emotion data, the input data being the specific data and emotion data, and the output data being the customized derived model, and the terminal performs operations to optimize and customize the model.

[0544] Step 7:

[0545] The terminal prepares the customized derived model so that it can be used in an offline environment. The input data is the customized derived model, and the output data is an offline-compatible model. The terminal performs an operation to convert the model into a form that can be executed offline.

[0546] Step 8:

[0547] The server saves the customized model and makes it available for distribution to other users. The input data is the customized derived model, and the output data is the distributable model. The server saves the model and performs operations to set up distribution.

[0548] Step 9:

[0549] The server tracks the usage of the customized model and records usage data. The input data is usage data by other users, and the output data is tracking data. The server monitors the use of the model and accumulates the data.

[0550] Step 10:

[0551] The server calculates rewards based on the tracking data and pays them to users who provide models. The input data is the tracking data, and the output data is the reward amount. The server performs the reward calculations and pays the providers.

[0552] This allows us to generate derived models specialized for specific tasks by utilizing the user's specific knowledge and emotional data, and to operate flexibly in offline environments.In addition, we can increase user motivation by providing rewards to users.

[0553] The specific processing unit 290 transmits the result of the specific processing to the smart device 14. In the smart device 14, the control unit 46A causes the output device 40 to output the result of the specific processing. The microphone 38B acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 38B to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[0554] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[0555] In the above embodiment, an example in which the specific process is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific process may be performed by the smart device 14.

[0556] [Second embodiment]

[0557] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0558] 3, the data processing system 210 includes the data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0559] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[0560] The smart glasses 214 include a computer 36, a microphone 238, a speaker 240, a camera 42, and a communication I / F 44. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, and the camera 42 are also connected to the bus 52.

[0561] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[0562] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0563] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[0564] Fig. 4 shows an example of the main functions of the data processing device 12 and the smart glasses 214. As shown in Fig. 4, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[0565] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[0566] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[0567] In the smart glasses 214, the reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[0568] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the smart glasses 214 will be referred to as the "terminal."

[0569] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models dedicated to specific tasks and customize them with users' unique knowledge and data. Furthermore, this system can be used offline and has the ability to provide rewards each time a customized model is used by other users.

[0570] Server-side processing

[0571] 1. Creating and serving the model

[0572] The server holds the underlying generative AI model. When a user requests a smaller derivative model optimized for a specific task (e.g., text generation in a specific domain), the server generates the derivative model according to the request and provides it to the user.

[0573] 2. Managing Customizations

[0574] The server provides a customization interface. Users upload their own specific knowledge and data, and the server receives the data and retrains or fine-tunes the specified derived model to suit the user's needs. The customized model is saved as a new version.

[0575] 3. Model Distribution and Reward Management

[0576] The server provides a function for distributing customized models to other users. Each time other users use a customized model, the server tracks the usage and calculates rewards based on the tracking data. Rewards are paid to the user who provided the customized model.

[0577] Terminal side processing

[0578] 1. Customizing the model

[0579] The user's device receives the derived model provided by the server and customizes it using local data (e.g., reviews of a particular product). The customization process is performed on the device, and the customized model is saved locally.

[0580] 2. Use in offline environments

[0581] The device has the ability to run customized models offline, for example, a product review generation model can run offline and perform tasks even when there is no internet connection.

[0582] User processing

[0583] 1. Requests and Customizations

[0584] A user requests a derived model for a specific task from the server, then customizes the model by uploading specific knowledge and data using the provided interface, and the customized model is stored locally on the user's computer.

[0585] 2. Use of Models and Receipt of Payments

[0586] Users can share their customized models with other users, and the server tracks usage and provides rewards each time other users use the model.

[0587] Specific examples

[0588] 1. Server

[0589] The server holds a basic model for "medical report generation." When User A requests a report generation model specialized for a specific medical condition, the server generates a derived model corresponding to the request and provides it to User A.

[0590] 2. Terminal

[0591] User A customizes the derived model on the device using local data acquired from a specific hospital. The customized model is saved on the device and can be used offline.

[0592] 3. Users

[0593] User A distributes a customized "condition-specific report generation model" online. When User B uses the model to generate a condition report, the server tracks its usage and provides a reward to User A.

[0594] In this way, a system is realized in which servers, terminals, and users can cooperate to efficiently and flexibly utilize generative AI models, creating, distributing, and using models optimized for specific tasks.

[0595] The processing flow will be explained below.

[0596] Server-side processing

[0597] Step 1:

[0598] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[0599] Step 2:

[0600] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[0601] Step 3:

[0602] The server provides the generated small derivative model to the user.

[0603] Step 4:

[0604] The server provides a customization interface where users can upload their own knowledge and data.

[0605] Step 5:

[0606] The server uses the received data to retrain or fine-tune the derived model to suit the user's needs.

[0607] Step 6:

[0608] The server saves the customized model as a new version.

[0609] Step 7:

[0610] The server configures the customized model for distribution to other users.

[0611] Step 8:

[0612] Each time another user uses the customized model, the server tracks that usage.

[0613] Step 9:

[0614] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[0615] Terminal side processing

[0616] Step 1:

[0617] The terminal receives the derived model provided by the server.

[0618] Step 2:

[0619] The user's device loads the specific data stored locally.

[0620] Step 3:

[0621] The device uses the provided model to retrain or fine-tune based on local data.

[0622] Step 4:

[0623] The device stores the customized model locally.

[0624] Step 5:

[0625] The device runs the customized model in an offline environment to perform tasks.

[0626] User processing

[0627] Step 1:

[0628] A user requests a derived model corresponding to a particular task from the server.

[0629] Step 2:

[0630] Users upload their own knowledge and data through a provided interface.

[0631] Step 3:

[0632] The user receives the customized model from the server and uses it.

[0633] Step 4:

[0634] Users can set up sharing of their customized models with other users.

[0635] Step 5:

[0636] See and receive rewards when other users use your customized model.

[0637] Specific examples

[0638] Server-side example

[0639] Step 1:

[0640] User A sends a request for a "medical condition report generation model."

[0641] Step 2:

[0642] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[0643] Step 3:

[0644] The server provides the generated "medical condition report generation model" to User A.

[0645] Terminal side example

[0646] Step 1:

[0647] User A's terminal receives the "medical condition report generation model" provided by the server.

[0648] Step 2:

[0649] User A's terminal reads the medical condition data acquired from a specific hospital.

[0650] Step 3:

[0651] The device retrains the provided model using the "medical condition data" and generates a customized "medical condition report generation model."

[0652] Step 4:

[0653] The customized model is saved on User A's device.

[0654] User-side example

[0655] Step 1:

[0656] User A sets up the customized model to be publicly available to other users.

[0657] Step 2:

[0658] User B uses the model to generate a medical report.

[0659] Step 3:

[0660] The server tracks User B's usage and provides a reward to User A.

[0661] Example 1

[0662] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0663] In order to quickly and individually respond to today's diverse user needs, it is desirable to use generative AI models optimized for specific tasks. However, existing systems make it difficult for users to customize models using their own knowledge and data, and it is not easy to use them in offline environments, share customized models, or calculate rewards. This hinders the efficient and flexible use of generative AI models.

[0664] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[0665] In this invention, the server includes: means for storing a base model; means for generating a compact derivative model optimized for a specific task; means for providing the compact derivative model based on a user request; means for receiving user-specific knowledge and data and customizing the model; means for saving the customized model; means for distributing the customized model to other users; means for tracking use of the customized model; means for calculating and providing a reward to the user based on the tracking data; means for using the model in an offline environment; means for customizing the model provided by the server in a local environment; means for a user to upload data through an interface and retrain the model based on the data; means for tracking usage of the generated customized model each time it is used by other users and calculating a reward; and means for providing the reward calculated based on the usage as credits to the user's account. This enables users to easily and flexibly customize, use, and share generative AI models tailored to their specific needs.

[0666] A "foundation model" is a large-scale generative AI model that can handle a variety of tasks.

[0667] A "small derivative model" is a small, specialized generative AI model that optimizes a base model for a specific task.

[0668] A "request" is a user's request for the generation of a derived model corresponding to a particular task.

[0669] "Customization" is the process of adapting an existing model using a user's specific knowledge and data.

[0670] "Saving" is the act of permanently recording the generated model data so that it can be accessed later.

[0671] "Distribution" is the act of providing a customized model to other users.

[0672] "Tracking" means monitoring and recording the usage and performance of a model.

[0673] "Reward" is the compensation received by a user who provides a customized model.

[0674] An "offline environment" is a situation where there is no internet connection.

[0675] A "local environment" is a system environment that runs on a user's terminal.

[0676] An "interface" is the means or tool by which a user interacts with a system.

[0677] "Retraining" is the process of updating an existing model with new data.

[0678] An "account" is an individual identification information that is used when a user is registered in the system and information such as rewards is managed.

[0679] "Credit" is a unit or form of reward used within the system.

[0680] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models optimized for specific tasks, which can then be customized with the user's specific knowledge and data. Additionally, this system can be used offline and has the ability to provide rewards each time the customized model is used by other users.

[0681] The server stores the base model, and when a user requests a small derived model optimized for a specific task (e.g., generating medical reports or product reviews), the server generates a derived model adapted to the specific task from the base model and provides it to the user. The server performs this procedure using a deep learning framework such as TensorFlow or PyTorch.

[0682] After receiving the derived model from the server, the user customizes it in their local environment (on their device). The user's device receives the derived model provided by the server and retrains the model using local data (specific medical records, product reviews, etc.). The retraining process is also performed using TensorFlow, PyTorch, etc. The customized model is saved in the user's local environment, allowing it to be used in offline environments.

[0683] The server receives the specific knowledge and data uploaded by the user and provides an interface for customization. Once the user customizes the model, the server saves the new version of the model and provides settings for distribution to other users. The server also tracks the usage of the customized model each time it is used by other users, calculates rewards based on the number and frequency of use, and adds them as credits to the user's account.

[0684] As a specific example, if a server holds a basic model for "medical report generation," when User A requests a report generation model specialized for a specific medical condition, the server generates a derived model according to the request and provides it to User A. User A customizes the derived model on his terminal using medical data obtained from a specific hospital and uses the customized model in offline environments. Furthermore, when User A sets up online distribution of the customized "medical condition-specific report generation model" and User B uses that model to generate a medical condition report, the server tracks its usage and provides User A with a reward.

[0685] An example prompt might be, "I would like to request an AI model to generate medical reports specific to the symptoms of rheumatoid arthritis patients. Furthermore, I would like to customize the model using specific hospital data."

[0686] In this way, by coordinating the functions of the server, terminal, and user, it becomes possible to efficiently and flexibly generate, use, and share generative AI models optimized for specific tasks.

[0687] The flow of the identification process in the first embodiment will be described with reference to FIG.

[0688] Step 1:

[0689] The server receives requests from users for derivative models corresponding to particular tasks.

[0690] Input: User request information (e.g., "Medical report generation model specialized for rheumatoid arthritis patients")

[0691] Specific operation: The user sends a request for a generative AI model to the server. The expected prompt is, "I would like to request an AI model that generates medical reports specific to the symptoms of rheumatoid arthritis patients."

[0692] Step 2:

[0693] The server uses the underlying generative AI model to generate smaller derivative models optimized for specific tasks.

[0694] Input: Base model stored on the server, user request information

[0695] Output: Smaller variants optimized for specific tasks

[0696] Specific operation: The server uses deep learning frameworks such as TensorFlow and PyTorch to generate derived models from the base model according to the request.

[0697] Step 3:

[0698] The server provides the generated derived model to the user.

[0699] Input: Generated small derivative model

[0700] Output: Derived models provided to users (via download link or API)

[0701] Specific operation: The server provides the derived model to the user by sending a download link via email or by providing an API endpoint.

[0702] Step 4:

[0703] The user's terminal receives the derived model provided by the server and customizes it with local data.

[0704] Input: Provided derived model, user's local data (e.g. medical records)

[0705] Output:Customized model

[0706] Specific operation: The user's device acquires local data (e.g., medical data obtained from a specific hospital) and retrains the model using TensorFlow or PyTorch.

[0707] Step 5:

[0708] The server manages the customization process using data provided by the user.

[0709] Input: User-uploaded specific data, customization interface

[0710] Output: Retrained customization model, notification of customization completion

[0711] Specific operation: The server receives the data uploaded by the user, performs the re-training process, and then saves the new version of the customized model.

[0712] Step 6:

[0713] The server provides the configuration to distribute customized models to other users.

[0714] Input: Customized model, distribution settings request

[0715] Output: Distribution settings to other users

[0716] Specific operation: The server receives a distribution configuration request from the user, performs the configuration process, and generates a public link and API endpoint.

[0717] Step 7:

[0718] The server tracks usage and calculates rewards each time the customized model is used by other users.

[0719] Input: Model usage tracking data

[0720] Output: Usage tracking record, calculated rewards

[0721] How it works: The server tracks usage of API calls, etc., and calculates rewards based on the number and frequency of usage.

[0722] Step 8:

[0723] The server provides the calculated reward as a credit to the user's account.

[0724] Input: Calculated reward

[0725] Output: Credits added to the user's account

[0726] Specific operation: The server adds the calculated reward to the user's account as credits, which the user can check on the dashboard etc.

[0727] Step 9:

[0728] The user terminal uses the customized model in an offline environment.

[0729] Input:Customized Model

[0730] Output: offline generated tasks (e.g. offline generated medical reports)

[0731] Specific operation: Users can use customized models, even without an internet connection, to generate medical reports, for example.

[0732] (Application example 1)

[0733] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0734] For autonomous vehicles, route optimization and emergency response to real-time road conditions and emergency situations are important. However, it is difficult to easily utilize the customized AI models required to meet these requirements in current systems. There is also a need for models that can be used in offline environments. Furthermore, there is a lack of functionality to track the usage of various customized models and provide appropriate rewards to users.

[0735] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[0736] In this invention, the server includes means for storing a base model, means for generating a compact derivative model optimized for a specific task, means for providing the compact derivative model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for using the model in an offline environment, and means for generating and customizing a derivative model optimized for a specific task using the generative AI model to support route optimization and emergency response in autonomous vehicles. This improves the accuracy of route optimization and emergency response in autonomous vehicles and significantly improves user convenience.

[0737] A "foundation model" is a fundamental AI model that plays a key role in the system and serves as the basis for derivative models optimized for specific tasks.

[0738] A "small derivative model" is a lightweight, efficient AI model generated from a base model and optimized for a specific task.

[0739] "User" refers to an individual or organization that utilizes the system to request and customize models optimized for specific tasks.

[0740] A "request" refers to an action in which a user asks the system to generate a derived model specialized for a specific task.

[0741] "Customization" is the process of adjusting a model using a user's specific knowledge and data to optimize it for their specific needs and requirements.

[0742] An "offline environment" refers to a state in which an Internet connection is not established, and the system operates independently of a network.

[0743] "Route optimization" refers to the calculations and processes that an autonomous vehicle performs to select the optimal route to its destination.

[0744] "Emergency response" refers to the process of providing appropriate handling and countermeasures for emergencies such as accidents and breakdowns.

[0745] A "generative AI model" refers to an artificial intelligence model that generates sentences, images, etc. based on large amounts of data.

[0746] "Tracking" refers to the act of tracking the use of a Customized Model and collecting Usage Data.

[0747] "Reward" refers to the compensation paid to the provider when the customized model is used by other users.

[0748] An "autonomous vehicle" refers to a vehicle that drives autonomously using sensors and artificial intelligence technology.

[0749] The present invention relates to a system for supporting route optimization and emergency response in an autonomous driving vehicle, and specific embodiments thereof will be described below.

[0750] Server-side processing

[0751] The server maintains the underlying model and generates smaller derivative models optimized for specific tasks. When a user submits data and a request, the server processes it using the following methods:

[0752] 1. Creating and serving the model

[0753] Using a base generative AI model (e.g., GPT-3), a derivative model is generated based on the user's request and provided to the user.

[0754] 2. Managing Customizations

[0755] It provides a customization interface and receives user-specific knowledge and data to retrain or fine-tune derived models, while also referencing real-time data from Google Maps API, Weather API, etc.

[0756] 3. Model Distribution and Reward Management

[0757] The usage of the customized model is tracked and rewards are calculated based on the usage data, and rewards are provided to the user who provided the customized model.

[0758] Terminal side processing

[0759] Users also have the ability to customize and save models locally on their devices that are optimized for specific tasks.

[0760] 1. Customizing the model

[0761] The user's device receives the derived model provided by the server and customizes it using local data (e.g., traffic information for a specific area). This customization process is performed on the device to optimize the performance of the generated AI model.

[0762] 2. Use in offline environments

[0763] The customized model can also work offline, allowing it to perform tasks even when there is no internet connection.

[0764] User processing

[0765] Users use the system to request and customize derivative models for specific tasks.

[0766] 1. Requests and Customizations

[0767] Users submit requests to the server and upload specific data to customize the model, generating prompts and entering the required data.

[0768] 2. Use of Models and Receipt of Payments

[0769] Users can share their customized models with other users and receive rewards each time they use the model.

[0770] Specific examples

[0771] When a user travels from Tokyo to Yokohama, the following process takes place:

[0772] 1. Example of route optimization

[0773] The user enters their current location "Tokyo" and their destination "Yokohama."

[0774] Provides real-time data on "heavy traffic" and "sunny" weather.

[0775] Example prompt: "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Generate the optimal route."

[0776] The generated optimal route is returned.

[0777] 2. Specific examples of emergency response

[0778] A user reports an emergency condition "accident" and specifies the location "Highway 246."

[0779] Example prompt: "Emergency situation: Accident. Location: Highway 246. Generate the appropriate response."

[0780] The generated contingency plan is provided.

[0781] In this way, the system of the present invention can improve the accuracy of route optimization and emergency response in autonomous vehicles, significantly improving user safety and convenience.

[0782] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[0783] Step 1:

[0784] A user opens an autonomous vehicle application and sends a request for route optimization or emergency response to the server. The inputs include current location, destination, and real-time data (e.g., traffic conditions, weather information). The server receives this request and generates a prompt for the underlying generative AI model (e.g., GPT-3).

[0785] Step 2:

[0786] The server queries the base generative AI model using the generated prompt. An example of a generated prompt is "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Please generate the optimal route." The prompt is given as input, and the optimal route is generated as output.

[0787] Step 3:

[0788] The server provides the generated optimal route to the user, which then returns the route information to the user's device. The user can review this route information and customize the model locally as needed. The generated route information is used as input and displayed on the user's device as output.

[0789] Step 4:

[0790] The user takes the route generated on the device and customizes it based on their specific circumstances, for example, modifying the route to take into account specific traffic patterns or personal driving preferences. User-specific data is used as input, and a customization model is generated as output.

[0791] Step 5:

[0792] The device stores the customized model and allows it to run offline if necessary. The input is the customized model and the output is the saved model. This allows the model to run even without an internet connection.

[0793] Step 6:

[0794] When a user encounters a situation requiring emergency response, the device sends an emergency request to the server. The input includes the type of emergency situation (e.g., accident) and location information. The server receives this request and generates a prompt.

[0795] Step 7:

[0796] The server generates emergency response measures based on the generated prompt text. An example of a prompt text is "Emergency situation: Accident Location: Highway 246. Please generate appropriate response measures." The prompt text is given as input, and an emergency response measure is generated as output.

[0797] Step 8:

[0798] The server returns the generated contingency plan to the user and displays it on the user's device. The user can take appropriate action based on this information. The generated contingency plan is used as input and displayed on the user's device as output.

[0799] Step 9:

[0800] The server tracks all customized models and their usage, and calculates rewards, using the tracking data as input and generating reward information as output, which is provided to the user.

[0801] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[0802] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks, can be customized with users' specific knowledge and data, and is combined with an emotion engine that recognizes users' emotions. This system can also be used in offline environments and has the function of providing rewards each time the customized model is used by other users.

[0803] Server-side processing

[0804] 1. Creating and serving the model

[0805] The server maintains the base model and receives requests from users for derived models for specific tasks (e.g., text generation in a specific field). Based on the request, the server generates a small derived model from the base model that is optimized for the specific task and provides it to the user.

[0806] 2. Customization management and emotion engine integration

[0807] The server provides a customization interface where users can upload their specific knowledge and data. It then uses an emotion engine to recognize the user's emotions and optimizes the customization model based on that data. The received data is then used to retrain and fine-tune the model to suit the user.

[0808] 3. Model Distribution and Reward Management

[0809] The server saves the customized model as a new version and configures it for distribution to other users. Each time another user uses the customized model, it tracks its usage, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[0810] Terminal side processing

[0811] 1. Customizing the model and using emotion data

[0812] The user's device receives the derived model provided by the server, reads the specific data (e.g., specific product review information) stored locally, and uses the emotion engine to obtain the user's emotion data and customize the model based on that data. The customized model is stored locally.

[0813] 2. Use in offline environments

[0814] The device has the ability to run customized models offline, allowing tasks to be performed even when there is no internet connection.

[0815] User processing

[0816] 1. Requests and Customizations

[0817] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided interface, and customize the models using emotion data obtained from the emotion engine.

[0818] 2. Use of Models and Receipt of Payments

[0819] Users can share their customized models with other users and receive rewards each time they use the model.

[0820] Specific examples

[0821] Server-side example

[0822] 1. Receiving a request and providing a model

[0823] When user A requests a "medical condition report generation model," the server receives the request, generates a derived model specialized for "medical condition report generation" from the base model, and provides it to user A.

[0824] 2. Customization management and emotion engine integration

[0825] User A uploads data obtained from a specific hospital, and the server uses an emotion engine to analyze User A's emotion data. Based on the analysis results, the derived model is retrained to generate a customized "condition report generation model."

[0826] Terminal side example

[0827] 1. Customizing the model

[0828] User A's device receives the "medical condition report generation model" provided by the server, reads the locally stored medical condition data, and uses the emotion engine to obtain User A's emotional data and optimize the model based on that data.

[0829] 2. Use in offline environments

[0830] User A's terminal executes the customized "medical condition report generation model" in an offline environment to generate a medical condition report.

[0831] User-side example

[0832] 1. Request and Data Upload

[0833] User A requests a "condition report generation model" from the server and uploads data from a specific hospital. Furthermore, the model is customized using data from the emotion engine.

[0834] 2. Use of Models and Receipt of Payments

[0835] User A sets up a customized "medical condition report generation model" to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[0836] In this way, a system is realized in which servers, devices, users, and emotion engines cooperate to efficiently and flexibly utilize generative AI models, and create, distribute, and use models optimized for specific tasks.

[0837] The processing flow will be explained below.

[0838] Server-side processing

[0839] Step 1:

[0840] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[0841] Step 2:

[0842] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[0843] Step 3:

[0844] The server provides the generated small derivative model to the user.

[0845] Step 4:

[0846] The server provides a customization interface where users can upload their own knowledge and data.

[0847] Step 5:

[0848] The server uses an emotion engine to collect user emotion data.

[0849] Step 6:

[0850] The server retrains or fine-tunes the derived model based on the user's specific data and emotional data.

[0851] Step 7:

[0852] The server saves the customized model as a new version.

[0853] Step 8:

[0854] The server configures the customized model for distribution to other users.

[0855] Step 9:

[0856] Each time another user uses the customized model, the server tracks that usage.

[0857] Step 10:

[0858] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[0859] Terminal side processing

[0860] Step 1:

[0861] The terminal receives the derived model provided by the server.

[0862] Step 2:

[0863] The user's device loads the specific data stored locally.

[0864] Step 3:

[0865] The terminal uses an emotion engine to obtain emotion data of the user.

[0866] Step 4:

[0867] The device uses the provided model to retrain or fine-tune based on local data and emotion data.

[0868] Step 5:

[0869] The device stores the customized model locally.

[0870] Step 6:

[0871] The device runs the customized model in an offline environment to perform tasks.

[0872] User processing

[0873] Step 1:

[0874] A user requests a derived model corresponding to a particular task from the server.

[0875] Step 2:

[0876] Users upload their own knowledge and data through a provided interface.

[0877] Step 3:

[0878] The user provides emotion data through the emotion engine.

[0879] Step 4:

[0880] The user receives the customized model from the server and uses it.

[0881] Step 5:

[0882] Users can set up sharing of their customized models with other users.

[0883] Step 6:

[0884] See and receive rewards when other users use your customized model.

[0885] Specific examples

[0886] Server-side example

[0887] Step 1:

[0888] User A sends a request for a "medical condition report generation model."

[0889] Step 2:

[0890] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[0891] Step 3:

[0892] The server provides the generated "medical condition report generation model" to User A.

[0893] Step 4:

[0894] The server provides a customization interface where User A uploads specific hospital data.

[0895] Step 5:

[0896] The server uses an emotion engine to collect emotion data of user A.

[0897] Step 6:

[0898] The server retrains the derived model based on the uploaded data and emotion data of user A.

[0899] Step 7:

[0900] The server stores the customized model and serves it as a new version.

[0901] Terminal side example

[0902] Step 1:

[0903] User A's terminal receives the "medical condition report generation model" provided by the server.

[0904] Step 2:

[0905] User A's terminal reads the medical condition data acquired from a specific hospital.

[0906] Step 3:

[0907] User A's device uses an emotion engine to acquire User A's emotion data.

[0908] Step 4:

[0909] The device retrains using medical condition data and emotion data to generate a customized model.

[0910] Step 5:

[0911] The customized model is saved on User A's device.

[0912] Step 6:

[0913] User A's terminal generates a medical condition report even in an offline environment.

[0914] User-side example

[0915] Step 1:

[0916] User A requests the "medical condition report generation model" from the server.

[0917] Step 2:

[0918] User A uploads hospital data through the server interface.

[0919] Step 3:

[0920] User A provides his / her own emotion data through the emotion engine.

[0921] Step 4:

[0922] User A receives the customized model from the server and uses it.

[0923] Step 5:

[0924] User A sets up the customized model to be publicly available to other users.

[0925] Step 6:

[0926] User B uses the model to generate a medical condition report.

[0927] Step 7:

[0928] The server tracks user B's usage and provides a reward to user A.

[0929] Example 2

[0930] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[0931] Conventional generative AI models are based on large datasets, making them less versatile for specific tasks and difficult to customize to meet individual needs. They also lack the ability to reflect user emotions in real time, making personalization based on user emotions difficult. Furthermore, they are limited to offline use, requiring a constant internet connection.

[0932] The specific processing by the specific processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a compact derived model optimized for a specific task, means for providing the compact derived model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for acquiring user emotion data using an emotion engine and using it for customization, and means for using the model in an offline environment. This enables the generation of a model tailored to individual needs for a specific task, personalization based on the user's emotions, and use in an offline environment.

[0933] A "foundation model" refers to a large-scale generative AI model, a general-purpose model that can handle a variety of tasks.

[0934] "Small derivative models" refer to lightweight models that are optimized for specific tasks from the base model.

[0935] A "user request" refers to a request from a user to generate a derived model corresponding to a particular task.

[0936] "Customization interface" refers to the operation screen and input means that allow users to upload their own specific knowledge and data and customize the model.

[0937] An "emotion engine" is a software module that acquires and analyzes user emotional data and optimizes the model based on that data.

[0938] "Offline environment" refers to a state where there is no internet connection, and refers to the ability to run models in such an environment.

[0939] "Tracking Data" refers to data (such as the number of uses and duration) when a customized model is used by other users.

[0940] "Remuneration" refers to the consideration paid to a user who provides a customized model in accordance with the use of the model.

[0941] "Local environment" refers to the user's device and its internal storage, an environment where data is stored without going through the Internet.

[0942] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks and can be customized with the user's specific knowledge and data. It also incorporates an emotion engine that recognizes the user's emotions, making it usable in offline environments. It also has a function that provides rewards each time the customized model is used by other users.

[0943] Server-side processing

[0944] The server first stores a large-scale base generative AI model (e.g., GPT-3). This base model is a general-purpose model that can handle a variety of tasks. When a user requests a derived model for a specific task (e.g., generating a medical diagnosis report), the server generates a small, task-optimized derived model from the base model and provides it to the user.

[0945] For example, when user A requests a "medical condition report generation model," the server generates a derived model specialized for "medical condition report generation" from the base model and provides it to user A.

[0946] The server then provides a customization interface, allowing users to upload their own specific data (e.g., data obtained from hospitals or expertise), and utilizes an emotion engine to analyze the user's emotion data and optimize the model based on this data, resulting in a more personalized model.

[0947] The customized model is saved in the database and managed as a new version. The server tracks the usage of this model by other users, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[0948] Terminal side processing

[0949] The device downloads the derived model provided by the server and stores it locally. It also reads specific data (e.g., specific product review information) stored locally on the device. Furthermore, the device uses an emotion engine to obtain user emotion data through input devices such as a camera or microphone, and customizes the derived model based on that data.

[0950] The device also has the ability to run customized models offline, allowing it to complete tasks even when there is no internet connection.

[0951] For example, User A's device locally stores a "medical condition report generation model" and optimizes it by reading local medical condition data and emotion data. This customized model can be used to generate medical condition reports even in offline environments.

[0952] User processing

[0953] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided customization interface, and further customize the models using emotion data obtained from the emotion engine to optimize their performance.

[0954] Users can share their customized models with other users and receive rewards each time they use the model. The server calculates and provides rewards to users based on tracking data.

[0955] Examples of concrete examples and prompts

[0956] For example, User A requests a "medical condition report generation model" and uploads data from a specific hospital to the server. Furthermore, he customizes the model using data from the emotion engine and sets it up to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[0957] Example prompt sentence:

[0958] Generate a medical report: "Generate a medical report based on the following data: [patient information]"

[0959] Product Review Analysis: "Analyze the following review and extract positive and negative comments. Review: [Product Review Text]"

[0960] As described above, this invention provides a flexible system that can be used in offline environments by using a base generation AI model to generate customized derivative models specialized for specific tasks and optimizing them using user emotional data and specific knowledge.

[0961] The flow of the identification process in the second embodiment will be described with reference to FIG.

[0962] Step 1: A user requests a derived model for a specific task.

[0963] Input: A task request sent by a user through a web interface or a dedicated application. A specific example is a medical condition report generation model.

[0964] Processing: The request information is sent to the server, which receives the request and starts processing according to the request contents.

[0965] Output: The server sends a confirmation message back to the user indicating that it has received the request.

[0966] Step 2: The server generates a derived model from the base model.

[0967] Input: A request for a specific task received from a user.

[0968] Processing: The server uses a large-scale generative AI model (e.g., GPT-3) as a base to generate a smaller derivative model optimized for a specific task based on the request. Here, the program sets parameters related to the specific task and calls the generative AI model to create the derivative model.

[0969] Output: The generated derived model file.

[0970] Step 3: The server provides the derived model to the user.

[0971] Input: The generated derived model file.

[0972] Processing: The server generates a download link to provide the model to the user and sends the link information to the user. The model file is also temporarily saved on the server.

[0973] Output: A notification message with a download link.

[0974] Step 4: User uploads data using a customization interface

[0975] Input: User specific data (e.g., hospital report data and expertise) and emotional data.

[0976] Processing: The user uploads their own data and emotion data to the server using the customization interface. The emotion data is retrieved by the emotion engine.

[0977] Output: Files of characteristic data and emotion data sent to the server.

[0978] Step 5: The server analyzes the emotion data and retrains the model

[0979] Input: User-uploaded characteristic and emotion data.

[0980] Processing: The emotion engine analyzes the uploaded emotion data and retrains and customizes the derived model based on the analysis results. Retraining is performed using a dedicated algorithm based on the base model and uploaded data.

[0981] Output: Customized derivative model.

[0982] Step 6: The server saves the customized model and prepares it for deployment.

[0983] Input: Customized derivative model.

[0984] Processing: Save the customized model as a new version in the database and configure it for distribution to other users.

[0985] Output: Saved model data and distribution settings information.

[0986] Step 7: Track when others use your model

[0987] Input: Use models customized by other users.

[0988] Processing: The server tracks the usage of the model (number of uses, duration of use, etc.). This process is achieved by recording usage logs in a database.

[0989] Output: Tracking data (usage log).

[0990] Step 8: The server calculates and provides the reward

[0991] Input: Tracking data (usage logs).

[0992] Processing: The server calculates the reward for the provider of the customized model based on the tracking data, and the calculated reward is deposited into the user's account.

[0993] Output: A record of the reward payment and a payment confirmation message.

[0994] Step 9: The device downloads the derivative model and saves it locally.

[0995] Input: Download link.

[0996] Processing: The user's device downloads the derived model using the download link provided by the server and stores it in local storage.

[0997] Output: Locally saved model file.

[0998] Step 10: The device loads local data and acquires emotion data

[0999] Input: Locally stored characteristic data and real-time emotion data.

[1000] Processing: The user's device reads the locally stored specific data and uses the emotion engine to obtain the user's real-time emotion data. Emotion data is collected using input devices such as cameras and microphones and analyzed.

[1001] Output: Parsed sentiment data and characteristic data.

[1002] Step 11: The device customizes the model based on the acquired data

[1003] Input: Feature data and emotion data.

[1004] Processing: The device uses this data to customize a derived model, adjusting the model parameters in the local environment to optimize it for the user's individual needs.

[1005] Output:Customized model.

[1006] Step 12: Run the customized model in an offline environment

[1007] Input:Customized model.

[1008] Processing: The user's device can execute customized models to accomplish specific tasks even in offline environments without an internet connection.

[1009] Output: The result of the task execution (e.g., the generated medical condition report).

[1010] (Application example 2)

[1011] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the smart glasses 214 will be referred to as a "terminal."

[1012] Conventional generative AI models have difficulty creating small models specifically tailored to specific tasks, and the results obtained by using general-purpose models are often inaccurate or insufficient. Furthermore, customization using user-specific knowledge and data is limited, and because emotional data is not taken into account, incentives and reward systems based on user emotions are not fully utilized. Furthermore, because they cannot be used in offline environments, flexible operation that does not rely on the Internet is required. To solve these issues, a more effective and flexibly customizable generative AI model system is needed.

[1013] The specific processing by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a small derived model optimized for a specific task, means for receiving user-specific knowledge and data and customizing the model, means for acquiring user emotional data and using it to customize the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing rewards to the user based on the tracking data, means for using the model in an offline environment, and means for making recommendations as a smartphone app based on the user's purchase history and emotional data. This makes it possible to generate a derived model specialized for a specific task by utilizing the user's specific knowledge and emotional data, and to operate flexibly in an offline environment. Furthermore, providing rewards to users can increase their motivation.

[1014] A "foundation model" is a large-scale model that is central to a generative AI system and serves as the basis for generating smaller derivative models that are specialized for specific tasks.

[1015] A "small derivative model" is a small generative AI model that is optimized from the base model to specialize in a specific task.

[1016] "Specific knowledge and data" refers to information and data sets that are unique to each individual user and are used to customize the model.

[1017] "Emotion data" is data that represents the user's emotional state and is acquired through the emotion engine.

[1018] "Customization" is the process of optimizing a generative AI model for a specific user or task based on specific knowledge, data, and emotional data.

[1019] "Saving" refers to storing a generated or customized model in digital storage for later use.

[1020] "Distribution" refers to providing a customized model to other users, and is done over a communications network.

[1021] "Usage Tracking" is the process of recording and monitoring how your customized model is being used.

[1022] A "reward" is an incentive paid to the provider of a customized model when the customized model is used by another user.

[1023] "Offline environment" refers to an environment without an internet connection, and the model is required to function in this environment.

[1024] "Recommendation" refers to recommending specific products or services based on a user's purchasing history and emotional data.

[1025] A "smartphone app" is application software that runs on a smartphone and performs specific functions.

[1026] This invention is a system based on a large-scale generative AI model that generates smaller derivative models specialized for specific tasks and customizes them with the user's specific knowledge and data. Furthermore, by combining it with an emotion engine that recognizes the user's emotions, it provides a personalized user experience. This system can also be used offline and has the ability to provide rewards each time the customized model is used by other users.

[1027] Overview

[1028] Server-side processing

[1029] 1. Generate and serve the model:

[1030] The server maintains the base model and generates smaller derivative models optimized for specific tasks in response to user requests. When a user makes a specific request, the server creates a corresponding derivative model based on that request and provides it to the user.

[1031] 2. Customization management and emotion engine integration:

[1032] The server provides a customization interface, allowing users to upload their own knowledge and data, and then uses an emotion engine to analyze the user's emotion data and optimize the model based on that data. Through retraining and fine-tuning, a personalized model is generated.

[1033] 3. Model Distribution and Reward Management:

[1034] The server stores and manages customized models, makes them available for distribution to other users, monitors model usage based on tracking data, and calculates and pays rewards to model providers.

[1035] Terminal side processing

[1036] 1. Customizing the model and using emotion data:

[1037] The user's device receives the derived model provided by the server, reads the specific data stored locally (e.g., purchase history), and uses the emotion engine to obtain the user's emotion data and use it to optimize the model.

[1038] 2. Offline use:

[1039] The device allows customized models to run offline, allowing tasks to be performed even when there is no internet connection.

[1040] User processing

[1041] 1. Requests and Customization:

[1042] Users request derived models for specific tasks from the server and upload their specific knowledge and data through the provided interface, and also utilize emotional data obtained from the emotion engine to customize the models.

[1043] 2. Use of Models and Receipt of Payments:

[1044] Users can share their customized models with other users and receive rewards each time their models are used.

[1045] Processing Description

[1046] server

[1047] The server uses the following hardware and software:

[1048] Hardware: High-performance server machine

[1049] Software: Emotion engine, AI model generation library, database management system

[1050] The server maintains the base model and generates derivative models specialized for specific tasks, provides a customization interface, optimizes the model based on user-specific and emotional data, and tracks the use of customized models and calculates rewards.

[1051] Terminal

[1052] The device uses the following hardware and software:

[1053] Hardware: Smartphone

[1054] Software: Emotion engine library, model optimization module

[1055] The device receives the derived model provided by the server and customizes it based on local data and emotion data, allowing it to perform tasks even in offline environments.

[1056] User

[1057] Users can request derived models for specific tasks, upload their specific knowledge and data to the server, and receive optimal recommendations using the customized models, leveraging sentiment data.

[1058] Specific examples

[1059] As an example of usage, consider a situation in which a user receives a recommendation for a specific product through a smartphone app.

[1060] Users can receive personalized recommendations based on their purchasing history and emotional data.

[1061] An example of a prompt sentence is, "Recommend the best products based on the user's purchasing history and current emotional state."

[1062] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1063] Step 1:

[1064] The server holds the foundation model and receives requests from users. The requests include a specific task the user needs (e.g., recommendations based on purchase history). The input data is the user's request, and the output data is a task instruction for generating a derived model. Based on the task instruction, the server prepares to generate a derived model from the foundation model.

[1065] Step 2:

[1066] The server generates a small derived model optimized for a specific task. The input data here are the base model and the user's request, and the output data is a derived model suitable for the specific task. The derived model is generated from the base model and provided to the user.

[1067] Step 3:

[1068] The terminal receives the derived model provided from the server. The input data is the provided derived model, and the output data is the derived model stored in the terminal. In this step, the terminal performs an operation of locally storing the derived model.

[1069] Step 4:

[1070] The user uploads specific knowledge and data (e.g., purchase history data) to the terminal. The input data is the user's specific data, and the output data is customized data stored locally. The terminal receives and stores the data from the user.

[1071] Step 5:

[1072] The terminal uses an emotion engine to obtain the user's current emotion data. The input data is the user's real-time emotional state, and the output data is emotion data. The emotion engine analyzes the user's emotion and generates the emotion data.

[1073] Step 6:

[1074] The terminal customizes the derived model based on the user's specific data and emotion data, the input data being the specific data and emotion data, and the output data being the customized derived model, and the terminal performs operations to optimize and customize the model.

[1075] Step 7:

[1076] The terminal prepares the customized derived model so that it can be used in an offline environment. The input data is the customized derived model, and the output data is an offline-compatible model. The terminal performs an operation to convert the model into a form that can be executed offline.

[1077] Step 8:

[1078] The server saves the customized model and makes it available for distribution to other users. The input data is the customized derived model, and the output data is the distributable model. The server saves the model and performs operations to set up distribution.

[1079] Step 9:

[1080] The server tracks the usage of the customized model and records usage data. The input data is usage data by other users, and the output data is tracking data. The server monitors the use of the model and accumulates the data.

[1081] Step 10:

[1082] The server calculates rewards based on the tracking data and pays them to users who provide models. The input data is the tracking data, and the output data is the reward amount. The server performs the reward calculations and pays the providers.

[1083] This allows us to generate derived models specialized for specific tasks by utilizing the user's specific knowledge and emotional data, and to operate flexibly in offline environments.In addition, we can increase user motivation by providing rewards to users.

[1084] The specific processing unit 290 transmits the result of the specific processing to the smart glasses 214. In the smart glasses 214, the control unit 46A causes the speaker 240 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1085] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1086] In the above embodiment, an example in which the specific processing is performed by the data processing device 12 has been given, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the smart glasses 214.

[1087] [Third embodiment]

[1088] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[1089] 5, the data processing system 310 includes the data processing device 12 and a headset type terminal 314. An example of the data processing device 12 is a server.

[1090] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1091] The headset type terminal 314 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a display 343. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the display 343 are also connected to the bus 52.

[1092] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1093] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1094] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 control the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1095] Fig. 6 shows an example of the main functions of the data processing device 12 and the headset type terminal 314. As shown in Fig. 6, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1096] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1097] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1098] In the headset type terminal 314, a reception output process is performed by the processor 46. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1099] Next, a description will be given of the identification process performed by the identification processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as the "server" and the headset type terminal 314 will be referred to as the "terminal."

[1100] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models dedicated to specific tasks and customize them with users' unique knowledge and data. Furthermore, this system can be used offline and has the ability to provide rewards each time a customized model is used by other users.

[1101] Server-side processing

[1102] 1. Creating and serving the model

[1103] The server holds the underlying generative AI model. When a user requests a smaller derivative model optimized for a specific task (e.g., text generation in a specific domain), the server generates the derivative model according to the request and provides it to the user.

[1104] 2. Managing Customizations

[1105] The server provides a customization interface. Users upload their own specific knowledge and data, and the server receives the data and retrains or fine-tunes the specified derived model to suit the user's needs. The customized model is saved as a new version.

[1106] 3. Model Distribution and Reward Management

[1107] The server provides a function for distributing customized models to other users. Each time other users use a customized model, the server tracks the usage and calculates rewards based on the tracking data. Rewards are paid to the user who provided the customized model.

[1108] Terminal side processing

[1109] 1. Customizing the model

[1110] The user's device receives the derived model provided by the server and customizes it using local data (e.g., reviews of a particular product). The customization process is performed on the device, and the customized model is saved locally.

[1111] 2. Use in offline environments

[1112] The device has the ability to run customized models offline, for example, a product review generation model can run offline and perform tasks even when there is no internet connection.

[1113] User processing

[1114] 1. Requests and Customizations

[1115] A user requests a derived model for a specific task from the server, then customizes the model by uploading specific knowledge and data using the provided interface, and the customized model is stored locally on the user's computer.

[1116] 2. Use of Models and Receipt of Payments

[1117] Users can share their customized models with other users, and the server tracks usage and provides rewards each time other users use the model.

[1118] Specific examples

[1119] 1. Server

[1120] The server holds a basic model for "medical report generation." When User A requests a report generation model specialized for a specific medical condition, the server generates a derived model corresponding to the request and provides it to User A.

[1121] 2. Terminal

[1122] User A customizes the derived model on the device using local data acquired from a specific hospital. The customized model is saved on the device and can be used offline.

[1123] 3. Users

[1124] User A distributes a customized "condition-specific report generation model" online. When User B uses the model to generate a condition report, the server tracks its usage and provides a reward to User A.

[1125] In this way, a system is realized in which servers, terminals, and users can cooperate to efficiently and flexibly utilize generative AI models, creating, distributing, and using models optimized for specific tasks.

[1126] The processing flow will be explained below.

[1127] Server-side processing

[1128] Step 1:

[1129] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[1130] Step 2:

[1131] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[1132] Step 3:

[1133] The server provides the generated small derivative model to the user.

[1134] Step 4:

[1135] The server provides a customization interface where users can upload their own knowledge and data.

[1136] Step 5:

[1137] The server uses the received data to retrain or fine-tune the derived model to suit the user's needs.

[1138] Step 6:

[1139] The server saves the customized model as a new version.

[1140] Step 7:

[1141] The server configures the customized model for distribution to other users.

[1142] Step 8:

[1143] Each time another user uses the customized model, the server tracks that usage.

[1144] Step 9:

[1145] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[1146] Terminal side processing

[1147] Step 1:

[1148] The terminal receives the derived model provided by the server.

[1149] Step 2:

[1150] The user's device loads the specific data stored locally.

[1151] Step 3:

[1152] The device uses the provided model to retrain or fine-tune based on local data.

[1153] Step 4:

[1154] The device stores the customized model locally.

[1155] Step 5:

[1156] The device runs the customized model in an offline environment to perform tasks.

[1157] User processing

[1158] Step 1:

[1159] A user requests a derived model corresponding to a particular task from the server.

[1160] Step 2:

[1161] Users upload their own knowledge and data through a provided interface.

[1162] Step 3:

[1163] The user receives the customized model from the server and uses it.

[1164] Step 4:

[1165] Users can set up sharing of their customized models with other users.

[1166] Step 5:

[1167] See and receive rewards when other users use your customized model.

[1168] Specific examples

[1169] Server-side example

[1170] Step 1:

[1171] User A sends a request for a "medical condition report generation model."

[1172] Step 2:

[1173] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[1174] Step 3:

[1175] The server provides the generated "medical condition report generation model" to User A.

[1176] Terminal side example

[1177] Step 1:

[1178] User A's terminal receives the "medical condition report generation model" provided by the server.

[1179] Step 2:

[1180] User A's terminal reads the medical condition data acquired from a specific hospital.

[1181] Step 3:

[1182] The device retrains the provided model using the "medical condition data" and generates a customized "medical condition report generation model."

[1183] Step 4:

[1184] The customized model is saved on User A's device.

[1185] User-side example

[1186] Step 1:

[1187] User A sets up the customized model to be publicly available to other users.

[1188] Step 2:

[1189] User B uses the model to generate a medical report.

[1190] Step 3:

[1191] The server tracks User B's usage and provides a reward to User A.

[1192] Example 1

[1193] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1194] In order to quickly and individually respond to today's diverse user needs, it is desirable to use generative AI models optimized for specific tasks. However, existing systems make it difficult for users to customize models using their own knowledge and data, and it is not easy to use them in offline environments, share customized models, or calculate rewards. This hinders the efficient and flexible use of generative AI models.

[1195] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1196] In this invention, the server includes: means for storing a base model; means for generating a compact derivative model optimized for a specific task; means for providing the compact derivative model based on a user request; means for receiving user-specific knowledge and data and customizing the model; means for saving the customized model; means for distributing the customized model to other users; means for tracking use of the customized model; means for calculating and providing a reward to the user based on the tracking data; means for using the model in an offline environment; means for customizing the model provided by the server in a local environment; means for a user to upload data through an interface and retrain the model based on the data; means for tracking usage of the generated customized model each time it is used by other users and calculating a reward; and means for providing the reward calculated based on the usage as credits to the user's account. This enables users to easily and flexibly customize, use, and share generative AI models tailored to their specific needs.

[1197] A "foundation model" is a large-scale generative AI model that can handle a variety of tasks.

[1198] A "small derivative model" is a small, specialized generative AI model that optimizes a base model for a specific task.

[1199] A "request" is a user's request for the generation of a derived model corresponding to a particular task.

[1200] "Customization" is the process of adapting an existing model using a user's specific knowledge and data.

[1201] "Saving" is the act of permanently recording the generated model data so that it can be accessed later.

[1202] "Distribution" is the act of providing a customized model to other users.

[1203] "Tracking" means monitoring and recording the usage and performance of a model.

[1204] "Reward" is the compensation received by a user who provides a customized model.

[1205] An "offline environment" is a situation where there is no internet connection.

[1206] A "local environment" is a system environment that runs on a user's terminal.

[1207] An "interface" is the means or tool by which a user interacts with a system.

[1208] "Retraining" is the process of updating an existing model with new data.

[1209] An "account" is an individual identification information that is used when a user is registered in the system and information such as rewards is managed.

[1210] "Credit" is a unit or form of reward used within the system.

[1211] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models optimized for specific tasks, which can then be customized with the user's specific knowledge and data. Additionally, this system can be used offline and has the ability to provide rewards each time the customized model is used by other users.

[1212] The server stores the base model, and when a user requests a small derived model optimized for a specific task (e.g., generating medical reports or product reviews), the server generates a derived model adapted to the specific task from the base model and provides it to the user. The server performs this procedure using a deep learning framework such as TensorFlow or PyTorch.

[1213] After receiving the derived model from the server, the user customizes it in their local environment (on their device). The user's device receives the derived model provided by the server and retrains the model using local data (specific medical records, product reviews, etc.). The retraining process is also performed using TensorFlow, PyTorch, etc. The customized model is saved in the user's local environment, allowing it to be used in offline environments.

[1214] The server receives the specific knowledge and data uploaded by the user and provides an interface for customization. Once the user customizes the model, the server saves the new version of the model and provides settings for distribution to other users. The server also tracks the usage of the customized model each time it is used by other users, calculates rewards based on the number and frequency of use, and adds them as credits to the user's account.

[1215] As a specific example, if a server holds a basic model for "medical report generation," when User A requests a report generation model specialized for a specific medical condition, the server generates a derived model according to the request and provides it to User A. User A customizes the derived model on his terminal using medical data obtained from a specific hospital and uses the customized model in offline environments. Furthermore, when User A sets up online distribution of the customized "medical condition-specific report generation model" and User B uses that model to generate a medical condition report, the server tracks its usage and provides User A with a reward.

[1216] An example prompt might be, "I would like to request an AI model to generate medical reports specific to the symptoms of rheumatoid arthritis patients. Furthermore, I would like to customize the model using specific hospital data."

[1217] In this way, by coordinating the functions of the server, terminal, and user, it becomes possible to efficiently and flexibly generate, use, and share generative AI models optimized for specific tasks.

[1218] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1219] Step 1:

[1220] The server receives requests from users for derivative models corresponding to particular tasks.

[1221] Input: User request information (e.g., "Medical report generation model specialized for rheumatoid arthritis patients")

[1222] Specific operation: The user sends a request for a generative AI model to the server. The expected prompt is, "I would like to request an AI model that generates medical reports specific to the symptoms of rheumatoid arthritis patients."

[1223] Step 2:

[1224] The server uses the underlying generative AI model to generate smaller derivative models optimized for specific tasks.

[1225] Input: Base model stored on the server, user request information

[1226] Output: Smaller variants optimized for specific tasks

[1227] Specific operation: The server uses deep learning frameworks such as TensorFlow and PyTorch to generate derived models from the base model according to the request.

[1228] Step 3:

[1229] The server provides the generated derived model to the user.

[1230] Input: Generated small derivative model

[1231] Output: Derived models provided to users (via download link or API)

[1232] Specific operation: The server provides the derived model to the user by sending a download link via email or by providing an API endpoint.

[1233] Step 4:

[1234] The user's terminal receives the derived model provided by the server and customizes it with local data.

[1235] Input: Provided derived model, user's local data (e.g. medical records)

[1236] Output:Customized model

[1237] Specific operation: The user's device acquires local data (e.g., medical data obtained from a specific hospital) and retrains the model using TensorFlow or PyTorch.

[1238] Step 5:

[1239] The server manages the customization process using data provided by the user.

[1240] Input: User-uploaded specific data, customization interface

[1241] Output: Retrained customization model, notification of customization completion

[1242] Specific operation: The server receives the data uploaded by the user, performs the re-training process, and then saves the new version of the customized model.

[1243] Step 6:

[1244] The server provides the configuration to distribute customized models to other users.

[1245] Input: Customized model, distribution settings request

[1246] Output: Distribution settings to other users

[1247] Specific operation: The server receives a distribution configuration request from the user, performs the configuration process, and generates a public link and API endpoint.

[1248] Step 7:

[1249] The server tracks usage and calculates rewards each time the customized model is used by other users.

[1250] Input: Model usage tracking data

[1251] Output: Usage tracking record, calculated rewards

[1252] How it works: The server tracks usage of API calls, etc., and calculates rewards based on the number and frequency of usage.

[1253] Step 8:

[1254] The server provides the calculated reward as a credit to the user's account.

[1255] Input: Calculated reward

[1256] Output: Credits added to the user's account

[1257] Specific operation: The server adds the calculated reward to the user's account as credits, which the user can check on the dashboard etc.

[1258] Step 9:

[1259] The user terminal uses the customized model in an offline environment.

[1260] Input:Customized Model

[1261] Output: offline generated tasks (e.g. offline generated medical reports)

[1262] Specific operation: Users can use customized models, even without an internet connection, to generate medical reports, for example.

[1263] (Application example 1)

[1264] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1265] For autonomous vehicles, route optimization and emergency response to real-time road conditions and emergency situations are important. However, it is difficult to easily utilize the customized AI models required to meet these requirements in current systems. There is also a need for models that can be used in offline environments. Furthermore, there is a lack of functionality to track the usage of various customized models and provide appropriate rewards to users.

[1266] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1267] In this invention, the server includes means for storing a base model, means for generating a compact derivative model optimized for a specific task, means for providing the compact derivative model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for using the model in an offline environment, and means for generating and customizing a derivative model optimized for a specific task using the generative AI model to support route optimization and emergency response in autonomous vehicles. This improves the accuracy of route optimization and emergency response in autonomous vehicles and significantly improves user convenience.

[1268] A "foundation model" is a fundamental AI model that plays a key role in the system and serves as the basis for derivative models optimized for specific tasks.

[1269] A "small derivative model" is a lightweight, efficient AI model generated from a base model and optimized for a specific task.

[1270] "User" refers to an individual or organization that utilizes the system to request and customize models optimized for specific tasks.

[1271] A "request" refers to an action in which a user asks the system to generate a derived model specialized for a specific task.

[1272] "Customization" is the process of adjusting a model using a user's specific knowledge and data to optimize it for their specific needs and requirements.

[1273] An "offline environment" refers to a state in which an Internet connection is not established, and the system operates independently of a network.

[1274] "Route optimization" refers to the calculations and processes that an autonomous vehicle performs to select the optimal route to its destination.

[1275] "Emergency response" refers to the process of providing appropriate handling and countermeasures for emergencies such as accidents and breakdowns.

[1276] A "generative AI model" refers to an artificial intelligence model that generates sentences, images, etc. based on large amounts of data.

[1277] "Tracking" refers to the act of tracking the use of a Customized Model and collecting Usage Data.

[1278] "Reward" refers to the compensation paid to the provider when the customized model is used by other users.

[1279] An "autonomous vehicle" refers to a vehicle that drives autonomously using sensors and artificial intelligence technology.

[1280] The present invention relates to a system for supporting route optimization and emergency response in an autonomous driving vehicle, and specific embodiments thereof will be described below.

[1281] Server-side processing

[1282] The server maintains the underlying model and generates smaller derivative models optimized for specific tasks. When a user submits data and a request, the server processes it using the following methods:

[1283] 1. Creating and serving the model

[1284] Using a base generative AI model (e.g., GPT-3), a derivative model is generated based on the user's request and provided to the user.

[1285] 2. Managing Customizations

[1286] It provides a customization interface and receives user-specific knowledge and data to retrain or fine-tune derived models, while also referencing real-time data from Google Maps API, Weather API, etc.

[1287] 3. Model Distribution and Reward Management

[1288] The usage of the customized model is tracked and rewards are calculated based on the usage data, and rewards are provided to the user who provided the customized model.

[1289] Terminal side processing

[1290] Users also have the ability to customize and save models locally on their devices that are optimized for specific tasks.

[1291] 1. Customizing the model

[1292] The user's device receives the derived model provided by the server and customizes it using local data (e.g., traffic information for a specific area). This customization process is performed on the device to optimize the performance of the generated AI model.

[1293] 2. Use in offline environments

[1294] The customized model can also work offline, allowing it to perform tasks even when there is no internet connection.

[1295] User processing

[1296] Users use the system to request and customize derivative models for specific tasks.

[1297] 1. Requests and Customizations

[1298] Users submit requests to the server and upload specific data to customize the model, generating prompts and entering the required data.

[1299] 2. Use of Models and Receipt of Payments

[1300] Users can share their customized models with other users and receive rewards each time they use the model.

[1301] Specific examples

[1302] When a user travels from Tokyo to Yokohama, the following process takes place:

[1303] 1. Example of route optimization

[1304] The user enters their current location "Tokyo" and their destination "Yokohama."

[1305] Provides real-time data on "heavy traffic" and "sunny" weather.

[1306] Example prompt: "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Generate the optimal route."

[1307] The generated optimal route is returned.

[1308] 2. Specific examples of emergency response

[1309] A user reports an emergency condition "accident" and specifies the location "Highway 246."

[1310] Example prompt: "Emergency situation: Accident. Location: Highway 246. Generate the appropriate response."

[1311] The generated contingency plan is provided.

[1312] In this way, the system of the present invention can improve the accuracy of route optimization and emergency response in autonomous vehicles, significantly improving user safety and convenience.

[1313] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1314] Step 1:

[1315] A user opens an autonomous vehicle application and sends a request for route optimization or emergency response to the server. The inputs include current location, destination, and real-time data (e.g., traffic conditions, weather information). The server receives this request and generates a prompt for the underlying generative AI model (e.g., GPT-3).

[1316] Step 2:

[1317] The server queries the base generative AI model using the generated prompt. An example of a generated prompt is "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Please generate the optimal route." The prompt is given as input, and the optimal route is generated as output.

[1318] Step 3:

[1319] The server provides the generated optimal route to the user, which then returns the route information to the user's device. The user can review this route information and customize the model locally as needed. The generated route information is used as input and displayed on the user's device as output.

[1320] Step 4:

[1321] The user takes the route generated on the device and customizes it based on their specific circumstances, for example, modifying the route to take into account specific traffic patterns or personal driving preferences. User-specific data is used as input, and a customization model is generated as output.

[1322] Step 5:

[1323] The device stores the customized model and allows it to run offline if necessary. The input is the customized model and the output is the saved model. This allows the model to run even without an internet connection.

[1324] Step 6:

[1325] When a user encounters a situation requiring emergency response, the device sends an emergency request to the server. The input includes the type of emergency situation (e.g., accident) and location information. The server receives this request and generates a prompt.

[1326] Step 7:

[1327] The server generates emergency response measures based on the generated prompt text. An example of a prompt text is "Emergency situation: Accident Location: Highway 246. Please generate appropriate response measures." The prompt text is given as input, and an emergency response measure is generated as output.

[1328] Step 8:

[1329] The server returns the generated contingency plan to the user and displays it on the user's device. The user can take appropriate action based on this information. The generated contingency plan is used as input and displayed on the user's device as output.

[1330] Step 9:

[1331] The server tracks all customized models and their usage, and calculates rewards, using the tracking data as input and generating reward information as output, which is provided to the user.

[1332] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1333] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks, can be customized with users' specific knowledge and data, and is combined with an emotion engine that recognizes users' emotions. This system can also be used in offline environments and has the function of providing rewards each time the customized model is used by other users.

[1334] Server-side processing

[1335] 1. Creating and serving the model

[1336] The server maintains the base model and receives requests from users for derived models for specific tasks (e.g., text generation in a specific field). Based on the request, the server generates a small derived model from the base model that is optimized for the specific task and provides it to the user.

[1337] 2. Customization management and emotion engine integration

[1338] The server provides a customization interface where users can upload their specific knowledge and data. It then uses an emotion engine to recognize the user's emotions and optimizes the customization model based on that data. The received data is then used to retrain and fine-tune the model to suit the user.

[1339] 3. Model Distribution and Reward Management

[1340] The server saves the customized model as a new version and configures it for distribution to other users. Each time another user uses the customized model, it tracks its usage, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[1341] Terminal side processing

[1342] 1. Customizing the model and using emotion data

[1343] The user's device receives the derived model provided by the server, reads the specific data (e.g., specific product review information) stored locally, and uses the emotion engine to obtain the user's emotion data and customize the model based on that data. The customized model is stored locally.

[1344] 2. Use in offline environments

[1345] The device has the ability to run customized models offline, allowing tasks to be performed even when there is no internet connection.

[1346] User processing

[1347] 1. Requests and Customizations

[1348] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided interface, and customize the models using emotion data obtained from the emotion engine.

[1349] 2. Use of Models and Receipt of Payments

[1350] Users can share their customized models with other users and receive rewards each time they use the model.

[1351] Specific examples

[1352] Server-side example

[1353] 1. Receiving a request and providing a model

[1354] When user A requests a "medical condition report generation model," the server receives the request, generates a derived model specialized for "medical condition report generation" from the base model, and provides it to user A.

[1355] 2. Customization management and emotion engine integration

[1356] User A uploads data obtained from a specific hospital, and the server uses an emotion engine to analyze User A's emotion data. Based on the analysis results, the derived model is retrained to generate a customized "condition report generation model."

[1357] Terminal side example

[1358] 1. Customizing the model

[1359] User A's device receives the "medical condition report generation model" provided by the server, reads the locally stored medical condition data, and uses the emotion engine to obtain User A's emotional data and optimize the model based on that data.

[1360] 2. Use in offline environments

[1361] User A's terminal executes the customized "medical condition report generation model" in an offline environment to generate a medical condition report.

[1362] User-side example

[1363] 1. Request and Data Upload

[1364] User A requests a "condition report generation model" from the server and uploads data from a specific hospital. Furthermore, the model is customized using data from the emotion engine.

[1365] 2. Use of Models and Receipt of Payments

[1366] User A sets up a customized "medical condition report generation model" to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[1367] In this way, a system is realized in which servers, devices, users, and emotion engines cooperate to efficiently and flexibly utilize generative AI models, and create, distribute, and use models optimized for specific tasks.

[1368] The processing flow will be explained below.

[1369] Server-side processing

[1370] Step 1:

[1371] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[1372] Step 2:

[1373] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[1374] Step 3:

[1375] The server provides the generated small derivative model to the user.

[1376] Step 4:

[1377] The server provides a customization interface where users can upload their own knowledge and data.

[1378] Step 5:

[1379] The server uses an emotion engine to collect user emotion data.

[1380] Step 6:

[1381] The server retrains or fine-tunes the derived model based on the user's specific data and emotional data.

[1382] Step 7:

[1383] The server saves the customized model as a new version.

[1384] Step 8:

[1385] The server configures the customized model for distribution to other users.

[1386] Step 9:

[1387] Each time another user uses the customized model, the server tracks that usage.

[1388] Step 10:

[1389] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[1390] Terminal side processing

[1391] Step 1:

[1392] The terminal receives the derived model provided by the server.

[1393] Step 2:

[1394] The user's device loads the specific data stored locally.

[1395] Step 3:

[1396] The terminal uses an emotion engine to obtain emotion data of the user.

[1397] Step 4:

[1398] The device uses the provided model to retrain or fine-tune based on local data and emotion data.

[1399] Step 5:

[1400] The device stores the customized model locally.

[1401] Step 6:

[1402] The device runs the customized model in an offline environment to perform tasks.

[1403] User processing

[1404] Step 1:

[1405] A user requests a derived model corresponding to a particular task from the server.

[1406] Step 2:

[1407] Users upload their own knowledge and data through a provided interface.

[1408] Step 3:

[1409] The user provides emotion data through the emotion engine.

[1410] Step 4:

[1411] The user receives the customized model from the server and uses it.

[1412] Step 5:

[1413] Users can set up sharing of their customized models with other users.

[1414] Step 6:

[1415] See and receive rewards when other users use your customized model.

[1416] Specific examples

[1417] Server-side example

[1418] Step 1:

[1419] User A sends a request for a "medical condition report generation model."

[1420] Step 2:

[1421] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[1422] Step 3:

[1423] The server provides the generated "medical condition report generation model" to User A.

[1424] Step 4:

[1425] The server provides a customization interface where User A uploads specific hospital data.

[1426] Step 5:

[1427] The server uses an emotion engine to collect emotion data of user A.

[1428] Step 6:

[1429] The server retrains the derived model based on the uploaded data and emotion data of user A.

[1430] Step 7:

[1431] The server stores the customized model and serves it as a new version.

[1432] Terminal side example

[1433] Step 1:

[1434] User A's terminal receives the "medical condition report generation model" provided by the server.

[1435] Step 2:

[1436] User A's terminal reads the medical condition data acquired from a specific hospital.

[1437] Step 3:

[1438] User A's device uses an emotion engine to acquire User A's emotion data.

[1439] Step 4:

[1440] The device retrains using medical condition data and emotion data to generate a customized model.

[1441] Step 5:

[1442] The customized model is saved on User A's device.

[1443] Step 6:

[1444] User A's terminal generates a medical condition report even in an offline environment.

[1445] User-side example

[1446] Step 1:

[1447] User A requests the "medical condition report generation model" from the server.

[1448] Step 2:

[1449] User A uploads hospital data through the server interface.

[1450] Step 3:

[1451] User A provides his / her own emotion data through the emotion engine.

[1452] Step 4:

[1453] User A receives the customized model from the server and uses it.

[1454] Step 5:

[1455] User A sets up the customized model to be publicly available to other users.

[1456] Step 6:

[1457] User B uses the model to generate a medical condition report.

[1458] Step 7:

[1459] The server tracks user B's usage and provides a reward to user A.

[1460] Example 2

[1461] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1462] Conventional generative AI models are based on large datasets, making them less versatile for specific tasks and difficult to customize to meet individual needs. They also lack the ability to reflect user emotions in real time, making personalization based on user emotions difficult. Furthermore, they are limited to offline use, requiring a constant internet connection.

[1463] The specific processing by the specific processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a compact derived model optimized for a specific task, means for providing the compact derived model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for acquiring user emotion data using an emotion engine and using it for customization, and means for using the model in an offline environment. This enables the generation of a model tailored to individual needs for a specific task, personalization based on the user's emotions, and use in an offline environment.

[1464] A "foundation model" refers to a large-scale generative AI model, a general-purpose model that can handle a variety of tasks.

[1465] "Small derivative models" refer to lightweight models that are optimized for specific tasks from the base model.

[1466] A "user request" refers to a request from a user to generate a derived model corresponding to a particular task.

[1467] "Customization interface" refers to the operation screen and input means that allow users to upload their own specific knowledge and data and customize the model.

[1468] An "emotion engine" is a software module that acquires and analyzes user emotional data and optimizes the model based on that data.

[1469] "Offline environment" refers to a state where there is no internet connection, and refers to the ability to run models in such an environment.

[1470] "Tracking Data" refers to data (such as the number of uses and duration) when a customized model is used by other users.

[1471] "Remuneration" refers to the consideration paid to a user who provides a customized model in accordance with the use of the model.

[1472] "Local environment" refers to the user's device and its internal storage, an environment where data is stored without going through the Internet.

[1473] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks and can be customized with the user's specific knowledge and data. It also incorporates an emotion engine that recognizes the user's emotions, making it usable in offline environments. It also has a function that provides rewards each time the customized model is used by other users.

[1474] Server-side processing

[1475] The server first stores a large-scale base generative AI model (e.g., GPT-3). This base model is a general-purpose model that can handle a variety of tasks. When a user requests a derived model for a specific task (e.g., generating a medical diagnosis report), the server generates a small, task-optimized derived model from the base model and provides it to the user.

[1476] For example, when user A requests a "medical condition report generation model," the server generates a derived model specialized for "medical condition report generation" from the base model and provides it to user A.

[1477] The server then provides a customization interface, allowing users to upload their own specific data (e.g., data obtained from hospitals or expertise), and utilizes an emotion engine to analyze the user's emotion data and optimize the model based on this data, resulting in a more personalized model.

[1478] The customized model is saved in the database and managed as a new version. The server tracks the usage of this model by other users, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[1479] Terminal side processing

[1480] The device downloads the derived model provided by the server and stores it locally. It also reads specific data (e.g., specific product review information) stored locally on the device. Furthermore, the device uses an emotion engine to obtain user emotion data through input devices such as a camera or microphone, and customizes the derived model based on that data.

[1481] The device also has the ability to run customized models offline, allowing it to complete tasks even when there is no internet connection.

[1482] For example, User A's device locally stores a "medical condition report generation model" and optimizes it by reading local medical condition data and emotion data. This customized model can be used to generate medical condition reports even in offline environments.

[1483] User processing

[1484] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided customization interface, and further customize the models using emotion data obtained from the emotion engine to optimize their performance.

[1485] Users can share their customized models with other users and receive rewards each time they use the model. The server calculates and provides rewards to users based on tracking data.

[1486] Examples of concrete examples and prompts

[1487] For example, User A requests a "medical condition report generation model" and uploads data from a specific hospital to the server. Furthermore, he customizes the model using data from the emotion engine and sets it up to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[1488] Example prompt sentence:

[1489] Generate a medical report: "Generate a medical report based on the following data: [patient information]"

[1490] Product Review Analysis: "Analyze the following review and extract positive and negative comments. Review: [Product Review Text]"

[1491] As described above, this invention provides a flexible system that can be used in offline environments by using a base generation AI model to generate customized derivative models specialized for specific tasks and optimizing them using user emotional data and specific knowledge.

[1492] The flow of the identification process in the second embodiment will be described with reference to FIG.

[1493] Step 1: A user requests a derived model for a specific task.

[1494] Input: A task request sent by a user through a web interface or a dedicated application. A specific example is a medical condition report generation model.

[1495] Processing: The request information is sent to the server, which receives the request and starts processing according to the request contents.

[1496] Output: The server sends a confirmation message back to the user indicating that it has received the request.

[1497] Step 2: The server generates a derived model from the base model.

[1498] Input: A request for a specific task received from a user.

[1499] Processing: The server uses a large-scale generative AI model (e.g., GPT-3) as a base to generate a smaller derivative model optimized for a specific task based on the request. Here, the program sets parameters related to the specific task and calls the generative AI model to create the derivative model.

[1500] Output: The generated derived model file.

[1501] Step 3: The server provides the derived model to the user.

[1502] Input: The generated derived model file.

[1503] Processing: The server generates a download link to provide the model to the user and sends the link information to the user. The model file is also temporarily saved on the server.

[1504] Output: A notification message with a download link.

[1505] Step 4: User uploads data using a customization interface

[1506] Input: User specific data (e.g., hospital report data and expertise) and emotional data.

[1507] Processing: The user uploads their own data and emotion data to the server using the customization interface. The emotion data is retrieved by the emotion engine.

[1508] Output: Files of characteristic data and emotion data sent to the server.

[1509] Step 5: The server analyzes the emotion data and retrains the model

[1510] Input: User-uploaded characteristic and emotion data.

[1511] Processing: The emotion engine analyzes the uploaded emotion data and retrains and customizes the derived model based on the analysis results. Retraining is performed using a dedicated algorithm based on the base model and uploaded data.

[1512] Output: Customized derivative model.

[1513] Step 6: The server saves the customized model and prepares it for deployment.

[1514] Input: Customized derivative model.

[1515] Processing: Save the customized model as a new version in the database and configure it for distribution to other users.

[1516] Output: Saved model data and distribution settings information.

[1517] Step 7: Track when others use your model

[1518] Input: Use models customized by other users.

[1519] Processing: The server tracks the usage of the model (number of uses, duration of use, etc.). This process is achieved by recording usage logs in a database.

[1520] Output: Tracking data (usage log).

[1521] Step 8: The server calculates and provides the reward

[1522] Input: Tracking data (usage logs).

[1523] Processing: The server calculates the reward for the provider of the customized model based on the tracking data, and the calculated reward is deposited into the user's account.

[1524] Output: A record of the reward payment and a payment confirmation message.

[1525] Step 9: The device downloads the derivative model and saves it locally.

[1526] Input: Download link.

[1527] Processing: The user's device downloads the derived model using the download link provided by the server and stores it in local storage.

[1528] Output: Locally saved model file.

[1529] Step 10: The device loads local data and acquires emotion data

[1530] Input: Locally stored characteristic data and real-time emotion data.

[1531] Processing: The user's device reads the locally stored specific data and uses the emotion engine to obtain the user's real-time emotion data. Emotion data is collected using input devices such as cameras and microphones and analyzed.

[1532] Output: Parsed sentiment data and characteristic data.

[1533] Step 11: The device customizes the model based on the acquired data

[1534] Input: Feature data and emotion data.

[1535] Processing: The device uses this data to customize a derived model, adjusting the model parameters in the local environment to optimize it for the user's individual needs.

[1536] Output:Customized model.

[1537] Step 12: Run the customized model in an offline environment

[1538] Input:Customized model.

[1539] Processing: The user's device can execute customized models to accomplish specific tasks even in offline environments without an internet connection.

[1540] Output: The result of the task execution (e.g., the generated medical condition report).

[1541] (Application example 2)

[1542] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the headset type terminal 314 will be referred to as a "terminal."

[1543] Conventional generative AI models have difficulty creating small models specifically tailored to specific tasks, and the results obtained by using general-purpose models are often inaccurate or insufficient. Furthermore, customization using user-specific knowledge and data is limited, and because emotional data is not taken into account, incentives and reward systems based on user emotions are not fully utilized. Furthermore, because they cannot be used in offline environments, flexible operation that does not rely on the Internet is required. To solve these issues, a more effective and flexibly customizable generative AI model system is needed.

[1544] The specific processing by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a small derived model optimized for a specific task, means for receiving user-specific knowledge and data and customizing the model, means for acquiring user emotional data and using it to customize the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing rewards to the user based on the tracking data, means for using the model in an offline environment, and means for making recommendations as a smartphone app based on the user's purchase history and emotional data. This makes it possible to generate a derived model specialized for a specific task by utilizing the user's specific knowledge and emotional data, and to operate flexibly in an offline environment. Furthermore, providing rewards to users can increase their motivation.

[1545] A "foundation model" is a large-scale model that is central to a generative AI system and serves as the basis for generating smaller derivative models that are specialized for specific tasks.

[1546] A "small derivative model" is a small generative AI model that is optimized from the base model to specialize in a specific task.

[1547] "Specific knowledge and data" refers to information and data sets that are unique to each individual user and are used to customize the model.

[1548] "Emotion data" is data that represents the user's emotional state and is acquired through the emotion engine.

[1549] "Customization" is the process of optimizing a generative AI model for a specific user or task based on specific knowledge, data, and emotional data.

[1550] "Saving" refers to storing a generated or customized model in digital storage for later use.

[1551] "Distribution" refers to providing a customized model to other users, and is done over a communications network.

[1552] "Usage Tracking" is the process of recording and monitoring how your customized model is being used.

[1553] A "reward" is an incentive paid to the provider of a customized model when the customized model is used by another user.

[1554] "Offline environment" refers to an environment without an internet connection, and the model is required to function in this environment.

[1555] "Recommendation" refers to recommending specific products or services based on a user's purchasing history and emotional data.

[1556] A "smartphone app" is application software that runs on a smartphone and performs specific functions.

[1557] This invention is a system based on a large-scale generative AI model that generates smaller derivative models specialized for specific tasks and customizes them with the user's specific knowledge and data. Furthermore, by combining it with an emotion engine that recognizes the user's emotions, it provides a personalized user experience. This system can also be used offline and has the ability to provide rewards each time the customized model is used by other users.

[1558] Overview

[1559] Server-side processing

[1560] 1. Generate and serve the model:

[1561] The server maintains the base model and generates smaller derivative models optimized for specific tasks in response to user requests. When a user makes a specific request, the server creates a corresponding derivative model based on that request and provides it to the user.

[1562] 2. Customization management and emotion engine integration:

[1563] The server provides a customization interface, allowing users to upload their own knowledge and data, and then uses an emotion engine to analyze the user's emotion data and optimize the model based on that data. Through retraining and fine-tuning, a personalized model is generated.

[1564] 3. Model Distribution and Reward Management:

[1565] The server stores and manages customized models, makes them available for distribution to other users, monitors model usage based on tracking data, and calculates and pays rewards to model providers.

[1566] Terminal side processing

[1567] 1. Customizing the model and using emotion data:

[1568] The user's device receives the derived model provided by the server, reads the specific data stored locally (e.g., purchase history), and uses the emotion engine to obtain the user's emotion data and use it to optimize the model.

[1569] 2. Offline use:

[1570] The device allows customized models to run offline, allowing tasks to be performed even when there is no internet connection.

[1571] User processing

[1572] 1. Requests and Customization:

[1573] Users request derived models for specific tasks from the server and upload their specific knowledge and data through the provided interface, and also utilize emotional data obtained from the emotion engine to customize the models.

[1574] 2. Use of Models and Receipt of Payments:

[1575] Users can share their customized models with other users and receive rewards each time their models are used.

[1576] Processing Description

[1577] server

[1578] The server uses the following hardware and software:

[1579] Hardware: High-performance server machine

[1580] Software: Emotion engine, AI model generation library, database management system

[1581] The server maintains the base model and generates derivative models specialized for specific tasks, provides a customization interface, optimizes the model based on user-specific and emotional data, and tracks the use of customized models and calculates rewards.

[1582] Terminal

[1583] The device uses the following hardware and software:

[1584] Hardware: Smartphone

[1585] Software: Emotion engine library, model optimization module

[1586] The device receives the derived model provided by the server and customizes it based on local data and emotion data, allowing it to perform tasks even in offline environments.

[1587] User

[1588] Users can request derived models for specific tasks, upload their specific knowledge and data to the server, and receive optimal recommendations using the customized models, leveraging sentiment data.

[1589] Specific examples

[1590] As an example of usage, consider a situation in which a user receives a recommendation for a specific product through a smartphone app.

[1591] Users can receive personalized recommendations based on their purchasing history and emotional data.

[1592] An example of a prompt sentence is, "Recommend the best products based on the user's purchasing history and current emotional state."

[1593] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[1594] Step 1:

[1595] The server holds the foundation model and receives requests from users. The requests include a specific task the user needs (e.g., recommendations based on purchase history). The input data is the user's request, and the output data is a task instruction for generating a derived model. Based on the task instruction, the server prepares to generate a derived model from the foundation model.

[1596] Step 2:

[1597] The server generates a small derived model optimized for a specific task. The input data here are the base model and the user's request, and the output data is a derived model suitable for the specific task. The derived model is generated from the base model and provided to the user.

[1598] Step 3:

[1599] The terminal receives the derived model provided from the server. The input data is the provided derived model, and the output data is the derived model stored in the terminal. In this step, the terminal performs an operation of locally storing the derived model.

[1600] Step 4:

[1601] The user uploads specific knowledge and data (e.g., purchase history data) to the terminal. The input data is the user's specific data, and the output data is customized data stored locally. The terminal receives and stores the data from the user.

[1602] Step 5:

[1603] The terminal uses an emotion engine to obtain the user's current emotion data. The input data is the user's real-time emotional state, and the output data is emotion data. The emotion engine analyzes the user's emotion and generates the emotion data.

[1604] Step 6:

[1605] The terminal customizes the derived model based on the user's specific data and emotion data, the input data being the specific data and emotion data, and the output data being the customized derived model, and the terminal performs operations to optimize and customize the model.

[1606] Step 7:

[1607] The terminal prepares the customized derived model so that it can be used in an offline environment. The input data is the customized derived model, and the output data is an offline-compatible model. The terminal performs an operation to convert the model into a form that can be executed offline.

[1608] Step 8:

[1609] The server saves the customized model and makes it available for distribution to other users. The input data is the customized derived model, and the output data is the distributable model. The server saves the model and performs operations to set up distribution.

[1610] Step 9:

[1611] The server tracks the usage of the customized model and records usage data. The input data is usage data by other users, and the output data is tracking data. The server monitors the use of the model and accumulates the data.

[1612] Step 10:

[1613] The server calculates rewards based on the tracking data and pays them to users who provide models. The input data is the tracking data, and the output data is the reward amount. The server performs the reward calculations and pays the providers.

[1614] This allows us to generate derived models specialized for specific tasks by utilizing the user's specific knowledge and emotional data, and to operate flexibly in offline environments.In addition, we can increase user motivation by providing rewards to users.

[1615] The specific processing unit 290 transmits the result of the specific processing to the headset type terminal 314. In the headset type terminal 314, the control unit 46A causes the speaker 240 and the display 343 to output the result of the specific processing. The microphone 238 acquires audio indicating a user input regarding the result of the specific processing. The control unit 46A transmits audio data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the audio data.

[1616] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[1617] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the headset type terminal 314.

[1618] [Fourth embodiment]

[1619] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[1620] 7, a data processing system 410 includes a data processing device 12 and a robot 414. An example of the data processing device 12 is a server.

[1621] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 is an example of a "computer" according to the technology of the present disclosure. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, the RAM 30, and the storage 32 are connected to a bus 34. The database 24 and the communication I / F 26 are also connected to the bus 34. The communication I / F 26 is connected to a network 54. Examples of the network 54 include a WAN (Wide Area Network) and / or a LAN (Local Area Network).

[1622] The robot 414 includes a computer 36, a microphone 238, a speaker 240, a camera 42, a communication I / F 44, and a control target 443. The computer 36 includes a processor 46, a RAM 48, and a storage 50. The processor 46, the RAM 48, and the storage 50 are connected to a bus 52. The microphone 238, the speaker 240, the camera 42, and the control target 443 are also connected to the bus 52.

[1623] The microphone 238 receives instructions and the like from the user 20 by receiving voice uttered by the user 20. The microphone 238 captures the voice uttered by the user 20, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio in accordance with instructions from the processor 46.

[1624] Camera 42 is a small digital camera equipped with an optical system including a lens, aperture, and shutter, and an imaging element such as a CMOS (Complementary Metal-Oxide-Semiconductor) image sensor or a CCD (Charge Coupled Device) image sensor, and captures images of the surroundings of user 20 (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[1625] The communication I / F 44 is connected to a network 54. The communication I / Fs 44 and 26 are responsible for the exchange of various information between the processor 46 and the processor 28 via the network 54. The exchange of various information between the processor 46 and the processor 28 using the communication I / Fs 44 and 26 is carried out in a secure state.

[1626] The control object 443 includes a display device, LEDs in the eyes, and motors for driving the arms, hands, and feet. The posture and gestures of the robot 414 are controlled by controlling the motors of the arms, hands, and feet. Some of the emotions of the robot 414 can be expressed by controlling these motors. In addition, the facial expressions of the robot 414 can also be expressed by controlling the light emission state of the LEDs in the eyes of the robot 414.

[1627] Fig. 8 shows an example of the main functions of the data processing device 12 and the robot 414. As shown in Fig. 8, in the data processing device 12, a specific process is performed by the processor 28. A specific process program 56 is stored in the storage 32.

[1628] The specific processing program 56 is an example of a "program" according to the technology of the present disclosure. The processor 28 reads the specific processing program 56 from the storage 32 and executes the read specific processing program 56 on the RAM 30. The specific processing is realized by the processor 28 operating as a specific processing unit 290 in accordance with the specific processing program 56 executed on the RAM 30.

[1629] The storage 32 stores a data generation model 58 and an emotion identification model 59. The data generation model 58 and the emotion identification model 59 are used by the identification processing unit 290.

[1630] In the robot 414, the processor 46 performs the reception output process. A reception output program 60 is stored in the storage 50. The processor 46 reads the reception output program 60 from the storage 50 and executes the read reception output program 60 on the RAM 48. The reception output process is realized by the processor 46 operating as the control unit 46A in accordance with the reception output program 60 executed on the RAM 48.

[1631] Next, a description will be given of the specific processing performed by the specific processing unit 290 of the data processing device 12. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1632] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models dedicated to specific tasks and customize them with users' unique knowledge and data. Furthermore, this system can be used offline and has the ability to provide rewards each time a customized model is used by other users.

[1633] Server-side processing

[1634] 1. Creating and serving the model

[1635] The server holds the underlying generative AI model. When a user requests a smaller derivative model optimized for a specific task (e.g., text generation in a specific domain), the server generates the derivative model according to the request and provides it to the user.

[1636] 2. Managing Customizations

[1637] The server provides a customization interface. Users upload their own specific knowledge and data, and the server receives the data and retrains or fine-tunes the specified derived model to suit the user's needs. The customized model is saved as a new version.

[1638] 3. Model Distribution and Reward Management

[1639] The server provides a function for distributing customized models to other users. Each time other users use a customized model, the server tracks the usage and calculates rewards based on the tracking data. Rewards are paid to the user who provided the customized model.

[1640] Terminal side processing

[1641] 1. Customizing the model

[1642] The user's device receives the derived model provided by the server and customizes it using local data (e.g., reviews of a particular product). The customization process is performed on the device, and the customized model is saved locally.

[1643] 2. Use in offline environments

[1644] The device has the ability to run customized models offline, for example, a product review generation model can run offline and perform tasks even when there is no internet connection.

[1645] User processing

[1646] 1. Requests and Customizations

[1647] A user requests a derived model for a specific task from the server, then customizes the model by uploading specific knowledge and data using the provided interface, and the customized model is stored locally on the user's computer.

[1648] 2. Use of Models and Receipt of Payments

[1649] Users can share their customized models with other users, and the server tracks usage and provides rewards each time other users use the model.

[1650] Specific examples

[1651] 1. Server

[1652] The server holds a basic model for "medical report generation." When User A requests a report generation model specialized for a specific medical condition, the server generates a derived model corresponding to the request and provides it to User A.

[1653] 2. Terminal

[1654] User A customizes the derived model on the device using local data acquired from a specific hospital. The customized model is saved on the device and can be used offline.

[1655] 3. Users

[1656] User A distributes a customized "condition-specific report generation model" online. When User B uses the model to generate a condition report, the server tracks its usage and provides a reward to User A.

[1657] In this way, a system is realized in which servers, terminals, and users can cooperate to efficiently and flexibly utilize generative AI models, creating, distributing, and using models optimized for specific tasks.

[1658] The processing flow will be explained below.

[1659] Server-side processing

[1660] Step 1:

[1661] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[1662] Step 2:

[1663] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[1664] Step 3:

[1665] The server provides the generated small derivative model to the user.

[1666] Step 4:

[1667] The server provides a customization interface where users can upload their own knowledge and data.

[1668] Step 5:

[1669] The server uses the received data to retrain or fine-tune the derived model to suit the user's needs.

[1670] Step 6:

[1671] The server saves the customized model as a new version.

[1672] Step 7:

[1673] The server configures the customized model for distribution to other users.

[1674] Step 8:

[1675] Each time another user uses the customized model, the server tracks that usage.

[1676] Step 9:

[1677] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[1678] Terminal side processing

[1679] Step 1:

[1680] The terminal receives the derived model provided by the server.

[1681] Step 2:

[1682] The user's device loads the specific data stored locally.

[1683] Step 3:

[1684] The device uses the provided model to retrain or fine-tune based on local data.

[1685] Step 4:

[1686] The device stores the customized model locally.

[1687] Step 5:

[1688] The device runs the customized model in an offline environment to perform tasks.

[1689] User processing

[1690] Step 1:

[1691] A user requests a derived model corresponding to a particular task from the server.

[1692] Step 2:

[1693] Users upload their own knowledge and data through a provided interface.

[1694] Step 3:

[1695] The user receives the customized model from the server and uses it.

[1696] Step 4:

[1697] Users can set up sharing of their customized models with other users.

[1698] Step 5:

[1699] See and receive rewards when other users use your customized model.

[1700] Specific examples

[1701] Server-side example

[1702] Step 1:

[1703] User A sends a request for a "medical condition report generation model."

[1704] Step 2:

[1705] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[1706] Step 3:

[1707] The server provides the generated "medical condition report generation model" to User A.

[1708] Terminal side example

[1709] Step 1:

[1710] User A's terminal receives the "medical condition report generation model" provided by the server.

[1711] Step 2:

[1712] User A's terminal reads the medical condition data acquired from a specific hospital.

[1713] Step 3:

[1714] The device retrains the provided model using the "medical condition data" and generates a customized "medical condition report generation model."

[1715] Step 4:

[1716] The customized model is saved on User A's device.

[1717] User-side example

[1718] Step 1:

[1719] User A sets up the customized model to be publicly available to other users.

[1720] Step 2:

[1721] User B uses the model to generate a medical report.

[1722] Step 3:

[1723] The server tracks User B's usage and provides a reward to User A.

[1724] Example 1

[1725] Next, a description will be given of Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1726] In order to quickly and individually respond to today's diverse user needs, it is desirable to use generative AI models optimized for specific tasks. However, existing systems make it difficult for users to customize models using their own knowledge and data, and it is not easy to use them in offline environments, share customized models, or calculate rewards. This hinders the efficient and flexible use of generative AI models.

[1727] The specific processing by the specific processing unit 290 of the data processing device 12 in the first embodiment is realized by the following means.

[1728] In this invention, the server includes: means for storing a base model; means for generating a compact derivative model optimized for a specific task; means for providing the compact derivative model based on a user request; means for receiving user-specific knowledge and data and customizing the model; means for saving the customized model; means for distributing the customized model to other users; means for tracking use of the customized model; means for calculating and providing a reward to the user based on the tracking data; means for using the model in an offline environment; means for customizing the model provided by the server in a local environment; means for a user to upload data through an interface and retrain the model based on the data; means for tracking usage of the generated customized model each time it is used by other users and calculating a reward; and means for providing the reward calculated based on the usage as credits to the user's account. This enables users to easily and flexibly customize, use, and share generative AI models tailored to their specific needs.

[1729] A "foundation model" is a large-scale generative AI model that can handle a variety of tasks.

[1730] A "small derivative model" is a small, specialized generative AI model that optimizes a base model for a specific task.

[1731] A "request" is a user's request for the generation of a derived model corresponding to a particular task.

[1732] "Customization" is the process of adapting an existing model using a user's specific knowledge and data.

[1733] "Saving" is the act of permanently recording the generated model data so that it can be accessed later.

[1734] "Distribution" is the act of providing a customized model to other users.

[1735] "Tracking" means monitoring and recording the usage and performance of a model.

[1736] "Reward" is the compensation received by a user who provides a customized model.

[1737] An "offline environment" is a situation where there is no internet connection.

[1738] A "local environment" is a system environment that runs on a user's terminal.

[1739] An "interface" is the means or tool by which a user interacts with a system.

[1740] "Retraining" is the process of updating an existing model with new data.

[1741] An "account" is an individual identification information that is used when a user is registered in the system and information such as rewards is managed.

[1742] "Credit" is a unit or form of reward used within the system.

[1743] This invention is a system that uses a large-scale generative AI model as a foundation to generate smaller derivative models optimized for specific tasks, which can then be customized with the user's specific knowledge and data. Additionally, this system can be used offline and has the ability to provide rewards each time the customized model is used by other users.

[1744] The server stores the base model, and when a user requests a small derived model optimized for a specific task (e.g., generating medical reports or product reviews), the server generates a derived model adapted to the specific task from the base model and provides it to the user. The server performs this procedure using a deep learning framework such as TensorFlow or PyTorch.

[1745] After receiving the derived model from the server, the user customizes it in their local environment (on their device). The user's device receives the derived model provided by the server and retrains the model using local data (specific medical records, product reviews, etc.). The retraining process is also performed using TensorFlow, PyTorch, etc. The customized model is saved in the user's local environment, allowing it to be used in offline environments.

[1746] The server receives the specific knowledge and data uploaded by the user and provides an interface for customization. Once the user customizes the model, the server saves the new version of the model and provides settings for distribution to other users. The server also tracks the usage of the customized model each time it is used by other users, calculates rewards based on the number and frequency of use, and adds them as credits to the user's account.

[1747] As a specific example, if a server holds a basic model for "medical report generation," when User A requests a report generation model specialized for a specific medical condition, the server generates a derived model according to the request and provides it to User A. User A customizes the derived model on his terminal using medical data obtained from a specific hospital and uses the customized model in offline environments. Furthermore, when User A sets up online distribution of the customized "medical condition-specific report generation model" and User B uses that model to generate a medical condition report, the server tracks its usage and provides User A with a reward.

[1748] An example prompt might be, "I would like to request an AI model to generate medical reports specific to the symptoms of rheumatoid arthritis patients. Furthermore, I would like to customize the model using specific hospital data."

[1749] In this way, by coordinating the functions of the server, terminal, and user, it becomes possible to efficiently and flexibly generate, use, and share generative AI models optimized for specific tasks.

[1750] The flow of the identification process in the first embodiment will be described with reference to FIG.

[1751] Step 1:

[1752] The server receives requests from users for derivative models corresponding to particular tasks.

[1753] Input: User request information (e.g., "Medical report generation model specialized for rheumatoid arthritis patients")

[1754] Specific operation: The user sends a request for a generative AI model to the server. The expected prompt is, "I would like to request an AI model that generates medical reports specific to the symptoms of rheumatoid arthritis patients."

[1755] Step 2:

[1756] The server uses the underlying generative AI model to generate smaller derivative models optimized for specific tasks.

[1757] Input: Base model stored on the server, user request information

[1758] Output: Smaller variants optimized for specific tasks

[1759] Specific operation: The server uses deep learning frameworks such as TensorFlow and PyTorch to generate derived models from the base model according to the request.

[1760] Step 3:

[1761] The server provides the generated derived model to the user.

[1762] Input: Generated small derivative model

[1763] Output: Derived models provided to users (via download link or API)

[1764] Specific operation: The server provides the derived model to the user by sending a download link via email or by providing an API endpoint.

[1765] Step 4:

[1766] The user's terminal receives the derived model provided by the server and customizes it with local data.

[1767] Input: Provided derived model, user's local data (e.g. medical records)

[1768] Output:Customized model

[1769] Specific operation: The user's device acquires local data (e.g., medical data obtained from a specific hospital) and retrains the model using TensorFlow or PyTorch.

[1770] Step 5:

[1771] The server manages the customization process using data provided by the user.

[1772] Input: User-uploaded specific data, customization interface

[1773] Output: Retrained customization model, notification of customization completion

[1774] Specific operation: The server receives the data uploaded by the user, performs the re-training process, and then saves the new version of the customized model.

[1775] Step 6:

[1776] The server provides the configuration to distribute customized models to other users.

[1777] Input: Customized model, distribution settings request

[1778] Output: Distribution settings to other users

[1779] Specific operation: The server receives a distribution configuration request from the user, performs the configuration process, and generates a public link and API endpoint.

[1780] Step 7:

[1781] The server tracks usage and calculates rewards each time the customized model is used by other users.

[1782] Input: Model usage tracking data

[1783] Output: Usage tracking record, calculated rewards

[1784] How it works: The server tracks usage of API calls, etc., and calculates rewards based on the number and frequency of usage.

[1785] Step 8:

[1786] The server provides the calculated reward as a credit to the user's account.

[1787] Input: Calculated reward

[1788] Output: Credits added to the user's account

[1789] Specific operation: The server adds the calculated reward to the user's account as credits, which the user can check on the dashboard etc.

[1790] Step 9:

[1791] The user terminal uses the customized model in an offline environment.

[1792] Input:Customized Model

[1793] Output: offline generated tasks (e.g. offline generated medical reports)

[1794] Specific operation: Users can use customized models, even without an internet connection, to generate medical reports, for example.

[1795] (Application example 1)

[1796] Next, a description will be given of Application Example 1. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1797] For autonomous vehicles, route optimization and emergency response to real-time road conditions and emergency situations are important. However, it is difficult to easily utilize the customized AI models required to meet these requirements in current systems. There is also a need for models that can be used in offline environments. Furthermore, there is a lack of functionality to track the usage of various customized models and provide appropriate rewards to users.

[1798] The specific processing by the specific processing unit 290 of the data processing device 12 in the application example 1 is realized by the following means.

[1799] In this invention, the server includes means for storing a base model, means for generating a compact derivative model optimized for a specific task, means for providing the compact derivative model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for using the model in an offline environment, and means for generating and customizing a derivative model optimized for a specific task using the generative AI model to support route optimization and emergency response in autonomous vehicles. This improves the accuracy of route optimization and emergency response in autonomous vehicles and significantly improves user convenience.

[1800] A "foundation model" is a fundamental AI model that plays a key role in the system and serves as the basis for derivative models optimized for specific tasks.

[1801] A "small derivative model" is a lightweight, efficient AI model generated from a base model and optimized for a specific task.

[1802] "User" refers to an individual or organization that utilizes the system to request and customize models optimized for specific tasks.

[1803] A "request" refers to an action in which a user asks the system to generate a derived model specialized for a specific task.

[1804] "Customization" is the process of adjusting a model using a user's specific knowledge and data to optimize it for their specific needs and requirements.

[1805] An "offline environment" refers to a state in which an Internet connection is not established, and the system operates independently of a network.

[1806] "Route optimization" refers to the calculations and processes that an autonomous vehicle performs to select the optimal route to its destination.

[1807] "Emergency response" refers to the process of providing appropriate handling and countermeasures for emergencies such as accidents and breakdowns.

[1808] A "generative AI model" refers to an artificial intelligence model that generates sentences, images, etc. based on large amounts of data.

[1809] "Tracking" refers to the act of tracking the use of a Customized Model and collecting Usage Data.

[1810] "Reward" refers to the compensation paid to the provider when the customized model is used by other users.

[1811] An "autonomous vehicle" refers to a vehicle that drives autonomously using sensors and artificial intelligence technology.

[1812] The present invention relates to a system for supporting route optimization and emergency response in an autonomous driving vehicle, and specific embodiments thereof will be described below.

[1813] Server-side processing

[1814] The server maintains the underlying model and generates smaller derivative models optimized for specific tasks. When a user submits data and a request, the server processes it using the following methods:

[1815] 1. Creating and serving the model

[1816] Using a base generative AI model (e.g., GPT-3), a derivative model is generated based on the user's request and provided to the user.

[1817] 2. Managing Customizations

[1818] It provides a customization interface and receives user-specific knowledge and data to retrain or fine-tune derived models, while also referencing real-time data from Google Maps API, Weather API, etc.

[1819] 3. Model Distribution and Reward Management

[1820] The usage of the customized model is tracked and rewards are calculated based on the usage data, and rewards are provided to the user who provided the customized model.

[1821] Terminal side processing

[1822] Users also have the ability to customize and save models locally on their devices that are optimized for specific tasks.

[1823] 1. Customizing the model

[1824] The user's device receives the derived model provided by the server and customizes it using local data (e.g., traffic information for a specific area). This customization process is performed on the device to optimize the performance of the generated AI model.

[1825] 2. Use in offline environments

[1826] The customized model can also work offline, allowing it to perform tasks even when there is no internet connection.

[1827] User processing

[1828] Users use the system to request and customize derivative models for specific tasks.

[1829] 1. Requests and Customizations

[1830] Users submit requests to the server and upload specific data to customize the model, generating prompts and entering the required data.

[1831] 2. Use of Models and Receipt of Payments

[1832] Users can share their customized models with other users and receive rewards each time they use the model.

[1833] Specific examples

[1834] When a user travels from Tokyo to Yokohama, the following process takes place:

[1835] 1. Example of route optimization

[1836] The user enters their current location "Tokyo" and their destination "Yokohama."

[1837] Provides real-time data on "heavy traffic" and "sunny" weather.

[1838] Example prompt: "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Generate the optimal route."

[1839] The generated optimal route is returned.

[1840] 2. Specific examples of emergency response

[1841] A user reports an emergency condition "accident" and specifies the location "Highway 246."

[1842] Example prompt: "Emergency situation: Accident. Location: Highway 246. Generate the appropriate response."

[1843] The generated contingency plan is provided.

[1844] In this way, the system of the present invention can improve the accuracy of route optimization and emergency response in autonomous vehicles, significantly improving user safety and convenience.

[1845] The flow of the specific processing in the application example 1 will be described with reference to FIG.

[1846] Step 1:

[1847] A user opens an autonomous vehicle application and sends a request for route optimization or emergency response to the server. The inputs include current location, destination, and real-time data (e.g., traffic conditions, weather information). The server receives this request and generates a prompt for the underlying generative AI model (e.g., GPT-3).

[1848] Step 2:

[1849] The server queries the base generative AI model using the generated prompt. An example of a generated prompt is "Current location: Tokyo, Destination: Yokohama, Real-time data: {traffic: 'heavy', weather: 'sunny'}. Please generate the optimal route." The prompt is given as input, and the optimal route is generated as output.

[1850] Step 3:

[1851] The server provides the generated optimal route to the user, which then returns the route information to the user's device. The user can review this route information and customize the model locally as needed. The generated route information is used as input and displayed on the user's device as output.

[1852] Step 4:

[1853] The user takes the route generated on the device and customizes it based on their specific circumstances, for example, modifying the route to take into account specific traffic patterns or personal driving preferences. User-specific data is used as input, and a customization model is generated as output.

[1854] Step 5:

[1855] The device stores the customized model and allows it to run offline if necessary. The input is the customized model and the output is the saved model. This allows the model to run even without an internet connection.

[1856] Step 6:

[1857] When a user encounters a situation requiring emergency response, the device sends an emergency request to the server. The input includes the type of emergency situation (e.g., accident) and location information. The server receives this request and generates a prompt.

[1858] Step 7:

[1859] The server generates emergency response measures based on the generated prompt text. An example of a prompt text is "Emergency situation: Accident Location: Highway 246. Please generate appropriate response measures." The prompt text is given as input, and an emergency response measure is generated as output.

[1860] Step 8:

[1861] The server returns the generated contingency plan to the user and displays it on the user's device. The user can take appropriate action based on this information. The generated contingency plan is used as input and displayed on the user's device as output.

[1862] Step 9:

[1863] The server tracks all customized models and their usage, and calculates rewards, using the tracking data as input and generating reward information as output, which is provided to the user.

[1864] Furthermore, an emotion engine that estimates the user's emotion may be further combined. That is, the identification processing unit 290 may estimate the user's emotion using the emotion identification model 59, and perform identification processing using the user's emotion.

[1865] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks, can be customized with users' specific knowledge and data, and is combined with an emotion engine that recognizes users' emotions. This system can also be used in offline environments and has the function of providing rewards each time the customized model is used by other users.

[1866] Server-side processing

[1867] 1. Creating and serving the model

[1868] The server maintains the base model and receives requests from users for derived models for specific tasks (e.g., text generation in a specific field). Based on the request, the server generates a small derived model from the base model that is optimized for the specific task and provides it to the user.

[1869] 2. Customization management and emotion engine integration

[1870] The server provides a customization interface where users can upload their specific knowledge and data. It then uses an emotion engine to recognize the user's emotions and optimizes the customization model based on that data. The received data is then used to retrain and fine-tune the model to suit the user.

[1871] 3. Model Distribution and Reward Management

[1872] The server saves the customized model as a new version and configures it for distribution to other users. Each time another user uses the customized model, it tracks its usage, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[1873] Terminal side processing

[1874] 1. Customizing the model and using emotion data

[1875] The user's device receives the derived model provided by the server, reads the specific data (e.g., specific product review information) stored locally, and uses the emotion engine to obtain the user's emotion data and customize the model based on that data. The customized model is stored locally.

[1876] 2. Use in offline environments

[1877] The device has the ability to run customized models offline, allowing tasks to be performed even when there is no internet connection.

[1878] User processing

[1879] 1. Requests and Customizations

[1880] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided interface, and customize the models using emotion data obtained from the emotion engine.

[1881] 2. Use of Models and Receipt of Payments

[1882] Users can share their customized models with other users and receive rewards each time they use the model.

[1883] Specific examples

[1884] Server-side example

[1885] 1. Receiving a request and providing a model

[1886] When user A requests a "medical condition report generation model," the server receives the request, generates a derived model specialized for "medical condition report generation" from the base model, and provides it to user A.

[1887] 2. Customization management and emotion engine integration

[1888] User A uploads data obtained from a specific hospital, and the server uses an emotion engine to analyze User A's emotion data. Based on the analysis results, the derived model is retrained to generate a customized "condition report generation model."

[1889] Terminal side example

[1890] 1. Customizing the model

[1891] User A's device receives the "medical condition report generation model" provided by the server, reads the locally stored medical condition data, and uses the emotion engine to obtain User A's emotional data and optimize the model based on that data.

[1892] 2. Use in offline environments

[1893] User A's terminal executes the customized "medical condition report generation model" in an offline environment to generate a medical condition report.

[1894] User-side example

[1895] 1. Request and Data Upload

[1896] User A requests a "condition report generation model" from the server and uploads data from a specific hospital. Furthermore, the model is customized using data from the emotion engine.

[1897] 2. Use of Models and Receipt of Payments

[1898] User A sets up a customized "medical condition report generation model" to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[1899] In this way, a system is realized in which servers, devices, users, and emotion engines cooperate to efficiently and flexibly utilize generative AI models, and create, distribute, and use models optimized for specific tasks.

[1900] The processing flow will be explained below.

[1901] Server-side processing

[1902] Step 1:

[1903] The server maintains the foundation model and receives requests from users for derived models for specific tasks.

[1904] Step 2:

[1905] Based on the request, the server generates a smaller derivative model from the base model that is optimized for the specific task.

[1906] Step 3:

[1907] The server provides the generated small derivative model to the user.

[1908] Step 4:

[1909] The server provides a customization interface where users can upload their own knowledge and data.

[1910] Step 5:

[1911] The server uses an emotion engine to collect user emotion data.

[1912] Step 6:

[1913] The server retrains or fine-tunes the derived model based on the user's specific data and emotional data.

[1914] Step 7:

[1915] The server saves the customized model as a new version.

[1916] Step 8:

[1917] The server configures the customized model for distribution to other users.

[1918] Step 9:

[1919] Each time another user uses the customized model, the server tracks that usage.

[1920] Step 10:

[1921] The server calculates rewards based on the tracking data and pays them to the users who provided the models.

[1922] Terminal side processing

[1923] Step 1:

[1924] The terminal receives the derived model provided by the server.

[1925] Step 2:

[1926] The user's device loads the specific data stored locally.

[1927] Step 3:

[1928] The terminal uses an emotion engine to obtain emotion data of the user.

[1929] Step 4:

[1930] The device uses the provided model to retrain or fine-tune based on local data and emotion data.

[1931] Step 5:

[1932] The device stores the customized model locally.

[1933] Step 6:

[1934] The device runs the customized model in an offline environment to perform tasks.

[1935] User processing

[1936] Step 1:

[1937] A user requests a derived model corresponding to a particular task from the server.

[1938] Step 2:

[1939] Users upload their own knowledge and data through a provided interface.

[1940] Step 3:

[1941] The user provides emotion data through the emotion engine.

[1942] Step 4:

[1943] The user receives the customized model from the server and uses it.

[1944] Step 5:

[1945] Users can set up sharing of their customized models with other users.

[1946] Step 6:

[1947] See and receive rewards when other users use your customized model.

[1948] Specific examples

[1949] Server-side example

[1950] Step 1:

[1951] User A sends a request for a "medical condition report generation model."

[1952] Step 2:

[1953] The server receives the request and generates a derived model from the base model that is specialized for "generating medical condition reports."

[1954] Step 3:

[1955] The server provides the generated "medical condition report generation model" to User A.

[1956] Step 4:

[1957] The server provides a customization interface where User A uploads specific hospital data.

[1958] Step 5:

[1959] The server uses an emotion engine to collect emotion data of user A.

[1960] Step 6:

[1961] The server retrains the derived model based on the uploaded data and emotion data of user A.

[1962] Step 7:

[1963] The server stores the customized model and serves it as a new version.

[1964] Terminal side example

[1965] Step 1:

[1966] User A's terminal receives the "medical condition report generation model" provided by the server.

[1967] Step 2:

[1968] User A's terminal reads the medical condition data acquired from a specific hospital.

[1969] Step 3:

[1970] User A's device uses an emotion engine to acquire User A's emotion data.

[1971] Step 4:

[1972] The device retrains using medical condition data and emotion data to generate a customized model.

[1973] Step 5:

[1974] The customized model is saved on User A's device.

[1975] Step 6:

[1976] User A's terminal generates a medical condition report even in an offline environment.

[1977] User-side example

[1978] Step 1:

[1979] User A requests the "medical condition report generation model" from the server.

[1980] Step 2:

[1981] User A uploads hospital data through the server interface.

[1982] Step 3:

[1983] User A provides his / her own emotion data through the emotion engine.

[1984] Step 4:

[1985] User A receives the customized model from the server and uses it.

[1986] Step 5:

[1987] User A sets up the customized model to be publicly available to other users.

[1988] Step 6:

[1989] User B uses the model to generate a medical condition report.

[1990] Step 7:

[1991] The server tracks user B's usage and provides a reward to user A.

[1992] Example 2

[1993] Next, a description will be given of Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[1994] Conventional generative AI models are based on large datasets, making them less versatile for specific tasks and difficult to customize to meet individual needs. They also lack the ability to reflect user emotions in real time, making personalization based on user emotions difficult. Furthermore, they are limited to offline use, requiring a constant internet connection.

[1995] The specific processing by the specific processing unit 290 of the data processing device 12 in Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a compact derived model optimized for a specific task, means for providing the compact derived model based on a user request, means for receiving user-specific knowledge and data and customizing the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing a reward to the user based on the tracking data, means for acquiring user emotion data using an emotion engine and using it for customization, and means for using the model in an offline environment. This enables the generation of a model tailored to individual needs for a specific task, personalization based on the user's emotions, and use in an offline environment.

[1996] A "foundation model" refers to a large-scale generative AI model, a general-purpose model that can handle a variety of tasks.

[1997] "Small derivative models" refer to lightweight models that are optimized for specific tasks from the base model.

[1998] A "user request" refers to a request from a user to generate a derived model corresponding to a particular task.

[1999] "Customization interface" refers to the operation screen and input means that allow users to upload their own specific knowledge and data and customize the model.

[2000] An "emotion engine" is a software module that acquires and analyzes user emotional data and optimizes the model based on that data.

[2001] "Offline environment" refers to a state where there is no internet connection, and refers to the ability to run models in such an environment.

[2002] "Tracking Data" refers to data (such as the number of uses and duration) when a customized model is used by other users.

[2003] "Remuneration" refers to the consideration paid to a user who provides a customized model in accordance with the use of the model.

[2004] "Local environment" refers to the user's device and its internal storage, an environment where data is stored without going through the Internet.

[2005] This invention is based on a large-scale generative AI model, and is a system that generates smaller derivative models specialized for specific tasks and can be customized with the user's specific knowledge and data. It also incorporates an emotion engine that recognizes the user's emotions, making it usable in offline environments. It also has a function that provides rewards each time the customized model is used by other users.

[2006] Server-side processing

[2007] The server first stores a large-scale base generative AI model (e.g., GPT-3). This base model is a general-purpose model that can handle a variety of tasks. When a user requests a derived model for a specific task (e.g., generating a medical diagnosis report), the server generates a small, task-optimized derived model from the base model and provides it to the user.

[2008] For example, when user A requests a "medical condition report generation model," the server generates a derived model specialized for "medical condition report generation" from the base model and provides it to user A.

[2009] The server then provides a customization interface, allowing users to upload their own specific data (e.g., data obtained from hospitals or expertise), and utilizes an emotion engine to analyze the user's emotion data and optimize the model based on this data, resulting in a more personalized model.

[2010] The customized model is saved in the database and managed as a new version. The server tracks the usage of this model by other users, calculates rewards based on the tracking data, and pays rewards to the user who provided the model.

[2011] Terminal side processing

[2012] The device downloads the derived model provided by the server and stores it locally. It also reads specific data (e.g., specific product review information) stored locally on the device. Furthermore, the device uses an emotion engine to obtain user emotion data through input devices such as a camera or microphone, and customizes the derived model based on that data.

[2013] The device also has the ability to run customized models offline, allowing it to complete tasks even when there is no internet connection.

[2014] For example, User A's device locally stores a "medical condition report generation model" and optimizes it by reading local medical condition data and emotion data. This customized model can be used to generate medical condition reports even in offline environments.

[2015] User processing

[2016] Users request derived models for specific tasks from the server, upload their specific knowledge and data through the provided customization interface, and further customize the models using emotion data obtained from the emotion engine to optimize their performance.

[2017] Users can share their customized models with other users and receive rewards each time they use the model. The server calculates and provides rewards to users based on tracking data.

[2018] Examples of concrete examples and prompts

[2019] For example, User A requests a "medical condition report generation model" and uploads data from a specific hospital to the server. Furthermore, he customizes the model using data from the emotion engine and sets it up to be shared with other users. When User B uses the model to generate a medical condition report, the server pays User A a reward based on the tracking data.

[2020] Example prompt sentence:

[2021] Generate a medical report: "Generate a medical report based on the following data: [patient information]"

[2022] Product Review Analysis: "Analyze the following review and extract positive and negative comments. Review: [Product Review Text]"

[2023] As described above, this invention provides a flexible system that can be used in offline environments by using a base generation AI model to generate customized derivative models specialized for specific tasks and optimizing them using user emotional data and specific knowledge.

[2024] The flow of the identification process in the second embodiment will be described with reference to FIG.

[2025] Step 1: A user requests a derived model for a specific task.

[2026] Input: A task request sent by a user through a web interface or a dedicated application. A specific example is a medical condition report generation model.

[2027] Processing: The request information is sent to the server, which receives the request and starts processing according to the request contents.

[2028] Output: The server sends a confirmation message back to the user indicating that it has received the request.

[2029] Step 2: The server generates a derived model from the base model.

[2030] Input: A request for a specific task received from a user.

[2031] Processing: The server uses a large-scale generative AI model (e.g., GPT-3) as a base to generate a smaller derivative model optimized for a specific task based on the request. Here, the program sets parameters related to the specific task and calls the generative AI model to create the derivative model.

[2032] Output: The generated derived model file.

[2033] Step 3: The server provides the derived model to the user.

[2034] Input: The generated derived model file.

[2035] Processing: The server generates a download link to provide the model to the user and sends the link information to the user. The model file is also temporarily saved on the server.

[2036] Output: A notification message with a download link.

[2037] Step 4: User uploads data using a customization interface

[2038] Input: User specific data (e.g., hospital report data and expertise) and emotional data.

[2039] Processing: The user uploads their own data and emotion data to the server using the customization interface. The emotion data is retrieved by the emotion engine.

[2040] Output: Files of characteristic data and emotion data sent to the server.

[2041] Step 5: The server analyzes the emotion data and retrains the model

[2042] Input: User-uploaded characteristic and emotion data.

[2043] Processing: The emotion engine analyzes the uploaded emotion data and retrains and customizes the derived model based on the analysis results. Retraining is performed using a dedicated algorithm based on the base model and uploaded data.

[2044] Output: Customized derivative model.

[2045] Step 6: The server saves the customized model and prepares it for deployment.

[2046] Input: Customized derivative model.

[2047] Processing: Save the customized model as a new version in the database and configure it for distribution to other users.

[2048] Output: Saved model data and distribution settings information.

[2049] Step 7: Track when others use your model

[2050] Input: Use models customized by other users.

[2051] Processing: The server tracks the usage of the model (number of uses, duration of use, etc.). This process is achieved by recording usage logs in a database.

[2052] Output: Tracking data (usage log).

[2053] Step 8: The server calculates and provides the reward

[2054] Input: Tracking data (usage logs).

[2055] Processing: The server calculates the reward for the provider of the customized model based on the tracking data, and the calculated reward is deposited into the user's account.

[2056] Output: A record of the reward payment and a payment confirmation message.

[2057] Step 9: The device downloads the derivative model and saves it locally.

[2058] Input: Download link.

[2059] Processing: The user's device downloads the derived model using the download link provided by the server and stores it in local storage.

[2060] Output: Locally saved model file.

[2061] Step 10: The device loads local data and acquires emotion data

[2062] Input: Locally stored characteristic data and real-time emotion data.

[2063] Processing: The user's device reads the locally stored specific data and uses the emotion engine to obtain the user's real-time emotion data. Emotion data is collected using input devices such as cameras and microphones and analyzed.

[2064] Output: Parsed sentiment data and characteristic data.

[2065] Step 11: The device customizes the model based on the acquired data

[2066] Input: Feature data and emotion data.

[2067] Processing: The device uses this data to customize a derived model, adjusting the model parameters in the local environment to optimize it for the user's individual needs.

[2068] Output:Customized model.

[2069] Step 12: Run the customized model in an offline environment

[2070] Input:Customized model.

[2071] Processing: The user's device can execute customized models to accomplish specific tasks even in offline environments without an internet connection.

[2072] Output: The result of the task execution (e.g., the generated medical condition report).

[2073] (Application example 2)

[2074] Next, a description will be given of Application Example 2. In the following description, the data processing device 12 will be referred to as a "server" and the robot 414 will be referred to as a "terminal."

[2075] Conventional generative AI models have difficulty creating small models specifically tailored to specific tasks, and the results obtained by using general-purpose models are often inaccurate or insufficient. Furthermore, customization using user-specific knowledge and data is limited, and because emotional data is not taken into account, incentives and reward systems based on user emotions are not fully utilized. Furthermore, because they cannot be used in offline environments, flexible operation that does not rely on the Internet is required. To solve these issues, a more effective and flexibly customizable generative AI model system is needed.

[2076] The specific processing by the specific processing unit 290 of the data processing device 12 in Application Example 2 is realized by the following means. In this invention, the server includes means for storing a base model, means for generating a small derived model optimized for a specific task, means for receiving user-specific knowledge and data and customizing the model, means for acquiring user emotional data and using it to customize the model, means for saving the customized model, means for distributing the customized model to other users, means for tracking use of the customized model, means for calculating and providing rewards to the user based on the tracking data, means for using the model in an offline environment, and means for making recommendations as a smartphone app based on the user's purchase history and emotional data. This makes it possible to generate a derived model specialized for a specific task by utilizing the user's specific knowledge and emotional data, and to operate flexibly in an offline environment. Furthermore, providing rewards to users can increase their motivation.

[2077] A "foundation model" is a large-scale model that is central to a generative AI system and serves as the basis for generating smaller derivative models that are specialized for specific tasks.

[2078] A "small derivative model" is a small generative AI model that is optimized from the base model to specialize in a specific task.

[2079] "Specific knowledge and data" refers to information and data sets that are unique to each individual user and are used to customize the model.

[2080] "Emotion data" is data that represents the user's emotional state and is acquired through the emotion engine.

[2081] "Customization" is the process of optimizing a generative AI model for a specific user or task based on specific knowledge, data, and emotional data.

[2082] "Saving" refers to storing a generated or customized model in digital storage for later use.

[2083] "Distribution" refers to providing a customized model to other users, and is done over a communications network.

[2084] "Usage Tracking" is the process of recording and monitoring how your customized model is being used.

[2085] A "reward" is an incentive paid to the provider of a customized model when the customized model is used by another user.

[2086] "Offline environment" refers to an environment without an internet connection, and the model is required to function in this environment.

[2087] "Recommendation" refers to recommending specific products or services based on a user's purchasing history and emotional data.

[2088] A "smartphone app" is application software that runs on a smartphone and performs specific functions.

[2089] This invention is a system based on a large-scale generative AI model that generates smaller derivative models specialized for specific tasks and customizes them with the user's specific knowledge and data. Furthermore, by combining it with an emotion engine that recognizes the user's emotions, it provides a personalized user experience. This system can also be used offline and has the ability to provide rewards each time the customized model is used by other users.

[2090] Overview

[2091] Server-side processing

[2092] 1. Generate and serve the model:

[2093] The server maintains the base model and generates smaller derivative models optimized for specific tasks in response to user requests. When a user makes a specific request, the server creates a corresponding derivative model based on that request and provides it to the user.

[2094] 2. Customization management and emotion engine integration:

[2095] The server provides a customization interface, allowing users to upload their own knowledge and data, and then uses an emotion engine to analyze the user's emotion data and optimize the model based on that data. Through retraining and fine-tuning, a personalized model is generated.

[2096] 3. Model Distribution and Reward Management:

[2097] The server stores and manages customized models, makes them available for distribution to other users, monitors model usage based on tracking data, and calculates and pays rewards to model providers.

[2098] Terminal side processing

[2099] 1. Customizing the model and using emotion data:

[2100] The user's device receives the derived model provided by the server, reads the specific data stored locally (e.g., purchase history), and uses the emotion engine to obtain the user's emotion data and use it to optimize the model.

[2101] 2. Offline use:

[2102] The device allows customized models to run offline, allowing tasks to be performed even when there is no internet connection.

[2103] User processing

[2104] 1. Requests and Customization:

[2105] Users request derived models for specific tasks from the server and upload their specific knowledge and data through the provided interface, and also utilize emotional data obtained from the emotion engine to customize the models.

[2106] 2. Use of Models and Receipt of Payments:

[2107] Users can share their customized models with other users and receive rewards each time their models are used.

[2108] Processing Description

[2109] server

[2110] The server uses the following hardware and software:

[2111] Hardware: High-performance server machine

[2112] Software: Emotion engine, AI model generation library, database management system

[2113] The server maintains the base model and generates derivative models specialized for specific tasks, provides a customization interface, optimizes the model based on user-specific and emotional data, and tracks the use of customized models and calculates rewards.

[2114] Terminal

[2115] The device uses the following hardware and software:

[2116] Hardware: Smartphone

[2117] Software: Emotion engine library, model optimization module

[2118] The device receives the derived model provided by the server and customizes it based on local data and emotion data, allowing it to perform tasks even in offline environments.

[2119] User

[2120] Users can request derived models for specific tasks, upload their specific knowledge and data to the server, and receive optimal recommendations using the customized models, leveraging sentiment data.

[2121] Specific examples

[2122] As an example of usage, consider a situation in which a user receives a recommendation for a specific product through a smartphone app.

[2123] Users can receive personalized recommendations based on their purchasing history and emotional data.

[2124] An example of a prompt sentence is, "Recommend the best products based on the user's purchasing history and current emotional state."

[2125] The flow of the specific processing in the application example 2 will be described with reference to FIG.

[2126] Step 1:

[2127] The server holds the foundation model and receives requests from users. The requests include a specific task the user needs (e.g., recommendations based on purchase history). The input data is the user's request, and the output data is a task instruction for generating a derived model. Based on the task instruction, the server prepares to generate a derived model from the foundation model.

[2128] Step 2:

[2129] The server generates a small derived model optimized for a specific task. The input data here are the base model and the user's request, and the output data is a derived model suitable for the specific task. The derived model is generated from the base model and provided to the user.

[2130] Step 3:

[2131] The terminal receives the derived model provided from the server. The input data is the provided derived model, and the output data is the derived model stored in the terminal. In this step, the terminal performs an operation of locally storing the derived model.

[2132] Step 4:

[2133] The user uploads specific knowledge and data (e.g., purchase history data) to the terminal. The input data is the user's specific data, and the output data is customized data stored locally. The terminal receives and stores the data from the user.

[2134] Step 5:

[2135] The terminal uses an emotion engine to obtain the user's current emotion data. The input data is the user's real-time emotional state, and the output data is emotion data. The emotion engine analyzes the user's emotion and generates the emotion data.

[2136] Step 6:

[2137] The terminal customizes the derived model based on the user's specific data and emotion data, the input data being the specific data and emotion data, and the output data being the customized derived model, and the terminal performs operations to optimize and customize the model.

[2138] Step 7:

[2139] The terminal prepares the customized derived model so that it can be used in an offline environment. The input data is the customized derived model, and the output data is an offline-compatible model. The terminal performs an operation to convert the model into a form that can be executed offline.

[2140] Step 8:

[2141] The server saves the customized model and makes it available for distribution to other users. The input data is the customized derived model, and the output data is the distributable model. The server saves the model and performs operations to set up distribution.

[2142] Step 9:

[2143] The server tracks the usage of the customized model and records usage data. The input data is usage data by other users, and the output data is tracking data. The server monitors the use of the model and accumulates the data.

[2144] Step 10:

[2145] The server calculates rewards based on the tracking data and pays them to users who provide models. The input data is the tracking data, and the output data is the reward amount. The server performs the reward calculations and pays the providers.

[2146] This allows us to generate derived models specialized for specific tasks by utilizing the user's specific knowledge and emotional data, and to operate flexibly in offline environments.In addition, we can increase user motivation by providing rewards to users.

[2147] The specific processing unit 290 transmits the result of the specific processing to the robot 414. In the robot 414, the control unit 46A causes the speaker 240 and the control target 443 to output the result of the specific processing. The microphone 238 acquires voice indicating a user input regarding the result of the specific processing. The control unit 46A transmits voice data indicating the user input acquired by the microphone 238 to the data processing device 12. In the data processing device 12, the specific processing unit 290 acquires the voice data.

[2148] The data generation model 58 is a so-called generative AI (Artificial Intelligence). An example of the data generation model 58 is ChatGPT (Internet Search<URL: https: / / openai.com / blog / chatgpt> ), Gemini (Internet search <url: https: gemini.google.com ?hl="ja">) and other generation AIs. The data generation model 58 is obtained by performing deep learning on a neural network. A prompt including an instruction is input to the data generation model 58, and inference data such as voice data indicating voice, text data indicating text, and image data indicating an image is also input. The data generation model 58 performs inference on the input inference data in accordance with the instruction indicated by the prompt, and outputs the inference result in a data format such as voice data and text data. Here, inference refers to, for example, analysis, classification, prediction, and / or summarization.

[2149] In the above embodiment, an example was given in which the specific processing is performed by the data processing device 12, but the technology of the present disclosure is not limited to this, and the specific processing may be performed by the robot 414.

[2150] The emotion identification model 59 as an emotion engine may determine the user's emotion according to a specific mapping. Specifically, the emotion identification model 59 may determine the user's emotion according to an emotion map (see FIG. 9), which is a specific mapping. Similarly, the emotion identification model 59 may determine the robot's emotion, and the identification processing unit 290 may perform identification processing using the robot's emotion.

[2151] FIG. 9 illustrates an emotion map 400 on which multiple emotions are mapped. In the emotion map 400, emotions are arranged in concentric circles radiating from the center. Emotions closer to the center of the concentric circles are more primitive. Emotions representing states and behaviors arising from a state of mind are arranged on the outer edges of the concentric circles. The concept of emotion includes both affect and mental states. Emotions generally generated from reactions occurring in the brain are arranged on the left side of the concentric circles. Emotions generally induced by situational judgment are arranged on the right side of the concentric circles. Emotions generally generated from reactions occurring in the brain and induced by situational judgment are arranged on the upper and lower sides of the concentric circles. Furthermore, the emotion of "pleasure" is arranged on the upper side of the concentric circles, and the emotion of "discomfort" is arranged on the lower side. In this way, in the emotion map 400, multiple emotions are mapped based on the structure by which emotions are generated, and emotions that tend to occur simultaneously are mapped close to each other.

[2152] These emotions are distributed in the 3 o'clock direction on emotion map 400, and typically fluctuate between relief and anxiety. In the right half of emotion map 400, situational awareness dominates over internal sensations, resulting in a sense of calm.

[2153] The inside of emotion map 400 represents what is going on in the mind, and the outside of emotion map 400 represents behavior, so the further you go outside emotion map 400, the more visible the emotions become (the more they are expressed in behavior).

[2154] Human emotions are based on various balances, such as posture and blood sugar levels. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. Emotions can also be created for robots, automobiles, and motorcycles, based on various balances, such as posture and remaining battery life. When these balances deviate from the ideal, a state of discomfort is indicated, and when they approach the ideal, a state of pleasure is indicated. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on Voice Emotion Recognition and Emotional Brain Physiological Signal Analysis Systems, Tokushima University, Doctoral Dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the "reaction" domain, where sensation is dominant. The right half of the emotion map lists emotions belonging to the "situation" domain, where situational awareness is dominant.

[2155] The emotion map defines two emotions that promote learning. One is a negative emotion on the situation side, around the middle of "repentance" or "reflection." In other words, this occurs when the robot experiences negative emotions such as "I never want to feel this way again" or "I don't want to be scolded again." The other is a positive emotion on the response side, around "desire." In other words, this occurs when the robot experiences positive feelings such as "I want more" or "I want to know more."

[2156] The emotion identification model 59 inputs user input into a pre-trained neural network, obtains emotion values ​​indicating each emotion shown in the emotion map 400, and determines the user's emotion. This neural network is pre-trained based on multiple pieces of training data that are combinations of user input and emotion values ​​indicating each emotion shown in the emotion map 400. Furthermore, this neural network is trained so that emotions that are located close to each other have similar values, as in the emotion map 900 shown in FIG. 10. FIG. 10 shows an example in which multiple emotions, "relieved," "calm," and "reassuring," have similar emotion values.

[2157] The system according to the present disclosure has been described above mainly with respect to the functions of the data processing device 12, but the system according to the present disclosure is not necessarily implemented on a server. The system according to the present disclosure may be implemented as a general information processing system. The present disclosure may be implemented, for example, as a software program running on a personal computer or an application running on a smartphone, etc. The method according to the present disclosure may be provided to users in the form of SaaS (Software as a Service).

[2158] In the above embodiment, an example was given in which the specific processing is performed by one computer 22, but the technology of the present disclosure is not limited to this, and the specific processing may be distributed and performed by a plurality of computers including the computer 22. For example, the data generation model 58 may be provided in an external device of the data processing device 12, and data may be generated in the external device in accordance with input data.

[2159] In the above embodiment, an example in which the specific processing program 56 is stored in the storage 32 has been described, but the technology of the present disclosure is not limited to this. For example, the specific processing program 56 may be stored in a portable, computer-readable, non-transitory storage medium such as a USB (Universal Serial Bus) memory. The specific processing program 56 stored in the non-transitory storage medium is installed in the computer 22 of the data processing device 12. The processor 28 executes the specific processing in accordance with the specific processing program 56.

[2160] Alternatively, the specific processing program 56 may be stored in a storage device such as a server connected to the data processing device 12 via the network 54, and the specific processing program 56 may be downloaded and installed on the computer 22 in response to a request from the data processing device 12.

[2161] It is not necessary to store all of the specific processing program 56 in a storage device such as a server connected to the data processing device 12 via the network 54, or to store all of the specific processing program 56 in the storage 32; only a portion of the specific processing program 56 may be stored.

[2162] The hardware resource for executing a specific process can be any of the following processors: An example of a processor is a CPU, which is a general-purpose processor that functions as a hardware resource for executing a specific process by executing software, i.e., a program. Another example of a processor is a dedicated electrical circuit, such as an FPGA (Field-Programmable Gate Array), a PLD (Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit), which is a processor with a circuit configuration designed specifically for executing a specific process. Each processor has built-in or connected memory, and each processor uses the memory to execute the specific process.

[2163] The hardware resource that executes the specific processing may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (for example, a combination of multiple FPGAs, or a combination of a CPU and an FPGA). Also, the hardware resource that executes the specific processing may be a single processor.

[2164] As an example of a system configured with a single processor, first, one processor is configured by combining one or more CPUs and software, and this processor functions as a hardware resource that executes a specific process. Second, there is a system that uses a processor that realizes the functions of an entire system including multiple hardware resources that execute a specific process on a single IC chip, as typified by SoC (System-on-a-chip). In this way, a specific process is realized using one or more of the above-mentioned various processors as hardware resources.

[2165] Furthermore, the hardware structure of these various processors can be, more specifically, an electric circuit that combines circuit elements such as semiconductor devices. The specific processing described above is merely an example. Therefore, it goes without saying that unnecessary steps may be deleted, new steps may be added, or the processing order may be rearranged, without departing from the spirit of the invention.

[2166] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[2167] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

[2168] The following is further disclosed regarding the above embodiment.

[2169] (Claim 1)

[2170] a means for maintaining the underlying model;

[2171] a means for generating smaller derivative models optimized for specific tasks;

[2172] means for providing a small derivative model based on a request from a user;

[2173] A means to customize the model by receiving user-specific knowledge and data;

[2174] a means for saving customized models;

[2175] a means for distributing the customized model to other users;

[2176] a means of tracking the use of the customized model; and

[2177] means for calculating and providing rewards to users based on the tracking data;

[2178] The system has a means for using the model in an offline environment.

[2179] (Claim 2)

[2180] 10. The system of claim 1, further comprising means for a user to customize the model in a local environment.

[2181] (Claim 3)

[2182] 10. The system of claim 1, further comprising: means for optimizing performance of the customized model.

[2183] (Claim 4)

[2184] 10. The system of claim 1, further comprising means for version control of the customized model.

[2185] (Claim 5)

[2186] 10. The system of claim 1, further comprising means for providing an interface for a user to share the customized model with other users.

[2187] "Example 1"

[2188] (Claim 1)

[2189] a means for maintaining the underlying model;

[2190] a means for generating smaller derivative models optimized for specific tasks;

[2191] means for providing a small derivative model based on a request from a user;

[2192] A means to customize the model by receiving user-specific knowledge and data;

[2193] a means for saving customized models;

[2194] a means for distributing the customized model to other users;

[2195] a means of tracking the use of the customized model; and

[2196] means for calculating and providing rewards to users based on the tracking data;

[2197] A means of using the model in an offline environment;

[2198] a means for locally customizing the model provided by the server;

[2199] a means for a user to upload data through the interface and retrain the model based on that data;

[2200] a means for tracking usage of the generated customized model and calculating rewards each time the customized model is used by other users;

[2201] means for providing a reward calculated based on usage as a credit to a user's account;

[2202] A system including:

[2203] (Claim 2)

[2204] 2. The system according to claim 1, further comprising means for customizing the derived model provided by the server on a user's terminal using local data.

[2205] (Claim 3)

[2206] 10. The system of claim 1, further comprising means for executing the customized model offline to perform certain tasks without an internet connection.

[2207] "Application Example 1"

[2208] (Claim 1)

[2209] a means for maintaining the underlying model;

[2210] a means for generating smaller derivative models optimized for specific tasks;

[2211] means for providing a small derivative model based on a request from a user;

[2212] A means to customize the model by receiving user-specific knowledge and data;

[2213] a means for saving customized models;

[2214] a means for distributing the customized model to other users;

[2215] a means of tracking the use of the customized model; and

[2216] means for calculating and providing rewards to users based on the tracking data;

[2217] A means of using the model in an offline environment;

[2218] A system that utilizes generative AI models to generate and customize derivative models optimized for specific tasks to support route optimization and emergency response in autonomous vehicles.

[2219] (Claim 2)

[2220] 10. The system of claim 1, further comprising means for a user to customize the model in a local environment.

[2221] (Claim 3)

[2222] 10. The system of claim 1, further comprising: means for optimizing performance of the customized model.

[2223] "Example 2: Combining Emotion Engines"

[2224] (Claim 1)

[2225] a means for maintaining the underlying model;

[2226] a means for generating smaller derivative models optimized for specific tasks;

[2227] means for providing a small derivative model based on a request from a user;

[2228] A means to customize the model by receiving user-specific knowledge and data;

[2229] a means for saving customized models;

[2230] a means for distributing the customized model to other users;

[2231] a means of tracking the use of the customized model; and

[2232] means for calculating and providing rewards to users based on the tracking data;

[2233] A means for acquiring user emotion data using an emotion engine and using the data for customization;

[2234] The system has a means for using the model in an offline environment.

[2235] (Claim 2)

[2236] 10. The system of claim 1, further comprising means for customizing the model in a local environment.

[2237] (Claim 3)

[2238] 10. The system of claim 1, further comprising: means for optimizing performance of the customized model.

[2239] "Application example 2 when combining emotion engines"

[2240] (Claim 1)

[2241] a means for maintaining the underlying model;

[2242] a means for generating smaller derivative models optimized for specific tasks;

[2243] means for providing a small derivative model based on a request from a user;

[2244] A means to customize the model by receiving user-specific knowledge and data;

[2245] A means of acquiring user emotion data and using it to customize the model;

[2246] a means for saving customized models;

[2247] a means for distributing the customized model to other users;

[2248] a means of tracking the use of the customized model; and

[2249] means for calculating and providing rewards to users based on the tracking data;

[2250] A means of using the model in an offline environment;

[2251] A system that has a means of making recommendations based on a user's purchasing history and emotional data as a smartphone app.

[2252] (Claim 2)

[2253] 10. The system of claim 1, further comprising means for a user to customize the model in a local environment.

[2254] (Claim 3)

[2255] 10. The system of claim 1, further comprising: means for optimizing performance of the customized model. [Explanation of symbols]

[2256] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot< / url:> < / url:> < / url:> < / url:>

Claims

1. a means for maintaining the underlying model; a means for generating smaller derivative models optimized for specific tasks; means for providing a small derivative model based on a request from a user; A means to customize the model by receiving user-specific knowledge and data; a means for saving customized models; a means for distributing the customized model to other users; a means of tracking the use of the customized model; and means for calculating and providing rewards to users based on the tracking data; The system has a means for using the model in an offline environment.

2. The system of claim 1 further comprising means for a user to customize the model in a local environment.

3. The system of claim 1 , further comprising means for optimizing performance of the customized model.

4. The system of claim 1 further comprising means for version control of the customized model.

5. The system of claim 1 , further comprising means for providing an interface for a user to share a customized model with other users.

Images