System

Abstract

An object of a system according to an embodiment is to provide an efficiently integrated system by utilizing a generation AI at each stage in creating a large-scale language model.SOLUTION: The specific processing unit 290 of the data processing system 12 in the system expands the datum by using the generation AI, and cleans the datum by using the generation AI. Further, the generative AI is utilized to generate the architecture and to optimize the training AI. In addition, a hyperparameter is tuned using a generative AI, automatic evaluation is performed using a generative AI, and fine adjustment is performed using a generative AI. In addition, the model is compressed using the generative AI, and continuous learning is performed using the generative AI.SELECTED DRAWING: Figure 2

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] Previous technology has faced the challenge of creating large-scale language models by lacking a system that efficiently integrates each stage, from data collection to model training and optimization.

[0005] The system according to the embodiment aims to provide an efficiently integrated system that utilizes generative AI at each stage in creating a large-scale language model. [Means for solving the problem]

[0006] The system according to the embodiment includes a data augmentation unit, a data cleaning unit, an architecture generation unit, a training data optimization unit, a hyperparameter tuning unit, an automatic evaluation unit, a fine-tuning unit, a model compression unit, and a continuous learning unit. The data augmentation unit uses a generation AI to augment data. The data cleaning unit uses a generation AI to clean data. The architecture generation unit uses a generation AI to generate an architecture. The training data optimization unit uses a generation AI to optimize training data. The hyperparameter tuning unit uses a generation AI to tune hyperparameters. The automatic evaluation unit uses a generation AI to perform automatic evaluation. The fine-tuning unit uses a generation AI to perform fine-tuning. The model compression unit uses a generation AI to compress a model. The continuous learning unit uses a generation AI to perform continuous learning. [Effects of the Invention]

[0007] The system according to the embodiment utilizes generative AI at each stage in creating a large-scale language model, and can provide an efficiently integrated system. [Brief explanation of the drawings]

[0008] [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. DETAILED DESCRIPTION OF THE INVENTION

[0009] 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.

[0010] First, the terms used in the following description will be explained.

[0011] 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, the 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), an APU (Accelerated Processing Unit), or a TPU (Tensor Processing Unit).

[0012] 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.

[0013] 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.

[0014] 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), and Bluetooth (registered trademark).

[0015] 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."

[0016] [First embodiment] FIG. 1 shows an example of the configuration of a data processing system 10 according to the first embodiment.

[0017] 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.

[0018] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, RAM 30, and 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).

[0019] 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.

[0020] The reception device 38 includes a touch panel 38A and a microphone 38B, and receives user input. The touch panel 38A detects contact with a pointer (for example, a pen or a finger) to receive user input by the touch of the pointer. 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 (see FIG. 2) acquires the data indicating the user input.

[0021] Output device 40 includes a display 40A and a speaker 40B, and presents data to a user by outputting the data in a form of expression that the user can perceive (e.g., audio and / or text). Display 40A displays visible information such as text and images in accordance with instructions from processor 46. Speaker 40B outputs audio in accordance with instructions from processor 46. 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.

[0022] 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.

[0023] FIG. 2 shows an example of the main functions of the data processing device 12 and the smart device 14.

[0024] 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.

[0025] 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. The identification processing unit 290 can estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0026] In the smart device 14, the specific processing is performed by the processor 46. The storage 50 stores a specific processing program 60. The specific processing program 60 is used together with the specific processing program 56 by the data processing system 10. The processor 46 reads the specific processing program 60 from the storage 50 and executes the read specific processing program 60 on the RAM 48. The specific processing is realized by the processor 46 operating as the control unit 46A in accordance with the specific processing program 60 executed on the RAM 48. Note that the smart device 14 may have a data generation model and an emotion identification model similar to the data generation model 58 and the emotion identification model 59.

[0027] Note that a device other than the data processing device 12 may have the data generation model 58. For example, a server device (e.g., a generation server) may have the data generation model 58. In this case, the data processing device 12 obtains a processing result (prediction result, etc.) using the data generation model 58 by communicating with the server device having the data generation model 58. Furthermore, the data processing device 12 may be a server device, or may be a terminal device owned by a user (e.g., a mobile phone, a robot, a home appliance, etc.). Next, an example of processing by the data processing system 10 according to the first embodiment will be described.

[0028] (Example 1) The integrated generative AI system according to an embodiment of the present invention is a system that streamlines the process of creating large-scale language models and maximizes model performance. As a result, the integrated generative AI system utilizes generative AI at each stage, from data collection to model training and optimization, to create efficient, high-performance large-scale language models.

[0029] The integrated generative AI system according to the embodiment includes a data augmentation unit, a data cleaning unit, an architecture generation unit, a training data optimization unit, a hyperparameter tuning unit, an automatic evaluation unit, a fine-tuning unit, a model compression unit, and a continuous learning unit. The data augmentation unit uses a generative AI to generate new data based on an existing dataset and expand the dataset. For example, the generative AI generates new sentences based on existing sentences to increase the diversity of the dataset. The generative AI receives inputs from prompts containing instructions for data augmentation, and the generative AI generates new data based on the prompts. The data cleaning unit uses a generative AI to automatically detect and clean noisy or inaccurate data in a dataset. For example, the generative AI detects and corrects typos and grammatical errors. The generative AI receives inputs from prompts containing instructions for data cleaning, and the generative AI cleans the data based on the prompts. The architecture generation unit uses a generative AI to automatically generate an optimal model architecture. For example, the generative AI designs the structure of a neural network optimal for a given task and proposes its architecture. The generative AI receives inputs from prompts containing instructions for architecture generation, and the generative AI generates an architecture based on the prompts. The training data optimization unit uses the generative AI to propose methods for optimizing training data. For example, the generative AI adjusts the balance of data to build an optimal dataset to improve model performance. The generative AI receives inputs from prompts containing instructions for optimizing training data, and the generative AI optimizes the data based on those prompts. The hyperparameter tuning unit uses the generative AI to automatically adjust the model's hyperparameters to find optimal settings. For example, the generative AI adjusts hyperparameters such as the learning rate and batch size to maximize model performance. The generative AI receives inputs from prompts containing instructions for tuning the hyperparameters, and the generative AI adjusts the hyperparameters based on those prompts. The automatic evaluation unit uses the generative AI to automatically evaluate the model's performance. For example, the generative AI uses test data to measure the model's precision and recall and reports the results.The input to the generative AI is a prompt containing instructions for model evaluation, and the generative AI evaluates the model based on the prompt. The fine-tuning unit uses the generative AI to fine-tune the model to improve its performance. For example, the generative AI performs fine-tuning to increase the adaptability of the model to a specific task. The input to the generative AI is a prompt containing instructions for fine-tuning, and the generative AI fine-tunes the model based on the prompt. The model compression unit uses the generative AI to compress the model size to enable efficient execution. For example, the generative AI reduces unnecessary parameters to make the model lighter. The input to the generative AI is a prompt containing instructions for model compression, and the generative AI compresses the model based on the prompt. The continuous learning unit uses the generative AI to help the model continuously learn from new data. For example, the generative AI incorporates new data and proposes a method for updating the model. The input to the generative AI is a prompt containing instructions for continuous learning, and the generative AI performs continuous learning based on the prompt. As a result, the integrated generative AI system according to the embodiment can streamline the process of creating large-scale language models and maximize model performance. For example, by building a high-quality dataset through data augmentation and cleaning, and training the model with optimal architecture and hyperparameters, it is possible to create a highly accurate language model. Furthermore, model compression and continuous learning enable efficient use in production environments.

[0030] The data augmentation unit can use the generative AI to generate data from different cultures and languages ​​and create a global dataset. The data augmentation unit, for example, uses the generative AI to generate text data from different cultures and languages. For example, it generates multilingual data such as English, Japanese, and French to build a global dataset. For example, to generate sentences from different cultures, the generative AI learns existing data from different cultures and generates new sentences based on their patterns. Also, for example, to generate sentences in different languages, the generative AI learns existing data from different languages ​​and generates new sentences based on their patterns. This allows the versatility of the model to be improved by creating a global dataset that includes data from different cultures and languages.

[0031] The data augmentation unit uses generative AI to generate data specialized for a specific industry or field of expertise, thereby building a highly specialized dataset. The data augmentation unit, for example, uses generative AI to generate text data specialized for a specific industry, such as medicine or law. For example, new text is generated based on medical papers or legal documents. The generative AI, for example, learns medical terminology and writing style and generates new medical-related text based on those patterns. The generative AI also learns legal terminology and writing style and generates new legal-related text based on those patterns. This makes it possible to create a highly specialized model by building a dataset specialized for a specific industry or field of expertise.

[0032] The data cleaning unit can use the generation AI to check the consistency of the data and automatically correct inconsistent data. The data cleaning unit, for example, uses the generation AI to check the consistency of text data and automatically correct inconsistent parts. For example, it corrects when different information is stated within the same context. The generation AI, for example, understands the context, detects inconsistent information, and corrects it to appropriate information. The generation AI also, for example, removes duplicate information and builds a consistent dataset. This makes it possible to maintain the consistency of the data and improve the reliability of the dataset.

[0033] The data cleaning unit can use the generation AI to evaluate the reliability of data and automatically remove low-reliability data. The data cleaning unit, for example, uses the generation AI to evaluate the reliability of text data and automatically remove low-reliability parts. For example, it deletes information whose source is unclear. The generation AI, for example, detects low-reliability information and removes that part from the dataset. The generation AI also, for example, selects only high-reliability information and builds a dataset. This makes it possible to improve the quality of the dataset by removing low-reliability data.

[0034] The data cleaning unit can use the generation AI to clean image and audio data and prepare a multimodal dataset. The data cleaning unit, for example, uses the generation AI to remove noise and improve the resolution of image data and clean it. For example, it can make blurry images clear. The generation AI, for example, evaluates the quality of image data and removes noise. The generation AI can also convert low-resolution images to high resolution. The generation AI can also clean audio data by removing noise and improving sound quality, for example. For example, it can remove background noise and make audio clear. In this way, cleaning image and audio data can improve the quality of the multimodal dataset.

[0035] The data cleaning unit can use the generation AI to perform data cleaning in real time and maintain a dynamic dataset. The data cleaning unit can, for example, use the generation AI to clean text data in real time and maintain a dynamic dataset. For example, it can instantly correct typos in response to user input. For example, the generation AI can check the consistency of the data in real time and correct contradictory information. In addition, the generation AI can, for example, evaluate the reliability of the data in real time and remove unreliable data. In this way, the quality of the dynamic dataset can be maintained by performing data cleaning in real time.

[0036] The architecture generation unit can use generative AI to automatically generate architectures that are optimal for different tasks and provide models optimized for each task. The architecture generation unit, for example, uses generative AI to automatically generate model architectures that are optimal for different tasks. For example, different neural network structures are proposed for image recognition tasks and natural language processing tasks. For example, the generative AI proposes a CNN architecture for image recognition tasks and an RNN architecture for natural language processing tasks. The generative AI also sets optimal hyperparameters for different tasks, for example, to maximize model performance. In this way, by automatically generating architectures that are optimal for different tasks, it is possible to provide models optimized for each task.

[0037] The architecture generation unit can use generative AI to generate a hybrid model that combines multiple architectures. The architecture generation unit, for example, uses generative AI to generate a hybrid model that combines different architectures. For example, it proposes a model that combines CNN and RNN. The generative AI generates a model that combines, for example, the feature extraction capabilities of CNN and the time series data processing capabilities of RNN. The generative AI also proposes a hybrid model that utilizes the strengths of different architectures, for example, to improve the performance of the model. In this way, by generating a hybrid model that combines multiple architectures, it is possible to provide a model that utilizes the strengths of each architecture.

[0038] The architecture generation unit uses generative AI to automatically generate architectures that correspond to image and audio data, and can build multimodal models. The architecture generation unit, for example, uses generative AI to automatically generate an optimal architecture that corresponds to image data. For example, it proposes a CNN architecture specialized for image classification tasks. The generative AI, for example, learns the features of the image data and generates an optimal architecture based on those patterns. The generative AI also automatically generates an optimal architecture that corresponds to audio data. For example, it proposes an RNN architecture specialized for audio recognition tasks. This makes it possible to build a multimodal model by automatically generating an architecture that corresponds to image and audio data.

[0039] The architecture generation unit can use a generative AI to generate an architecture in real time and provide a dynamic model. The architecture generation unit can use, for example, a generative AI to generate a model architecture in real time and provide a dynamic model. For example, the architecture generation unit can instantly propose an optimal architecture in response to user input. For example, the generative AI can analyze data in real time and generate an architecture based on the results. Furthermore, the generative AI can adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by generating an architecture in real time.

[0040] The training data optimization unit can use the generation AI to adjust the balance of the data and build a dataset that is not biased towards a particular class. The training data optimization unit, for example, uses the generation AI to adjust the balance of the training data and build a dataset that is not biased towards a particular class. For example, it equalizes the amount of data for each class. The generation AI, for example, analyzes the distribution of the data and builds a balanced dataset. In addition, the generation AI, for example, avoids data duplication in order to maintain data balance. In this way, by adjusting the balance of the data, it is possible to build a dataset that is not biased towards a particular class.

[0041] The training data optimization unit can use the generative AI to collect and integrate data from different data sources to increase data diversity. The training data optimization unit, for example, uses the generative AI to collect and integrate data from different data sources to increase data diversity. For example, it collects news articles and blog articles. The generative AI, for example, collects data from different data sources and integrates the data to build a diverse dataset. The generative AI also, for example, integrates data in different formats to increase the diversity of the dataset. This makes it possible to increase data diversity by collecting and integrating data from different data sources.

[0042] The training data optimization unit can use the generative AI to optimize training data for image and audio data and train a multimodal model. The training data optimization unit, for example, uses the generative AI to optimize training data for image data and train a multimodal model. For example, it builds a dataset that is optimal for an image classification task. For example, the generative AI analyzes the features of the image data and builds an optimal training dataset based on the patterns. Furthermore, the generative AI can optimize training data for audio data and build an optimal dataset for a speech recognition task. In this way, it is possible to train a multimodal model by optimizing the training data for image and audio data.

[0043] The training data optimization unit can use the generative AI to optimize the training data in real time and train a dynamic model. The training data optimization unit can use, for example, the generative AI to optimize the training data in real time and train a dynamic model. For example, the generative AI can instantly update the dataset in response to user input. For example, the generative AI can analyze data in real time and optimize the training data based on the results. In addition, the generative AI can adjust the balance of the data in real time and build a dataset that is not biased toward a particular class. This makes it possible to train a dynamic model by optimizing the training data in real time.

[0044] The hyperparameter tuning unit can use the generation AI to automatically adjust hyperparameters that are optimal for different tasks and provide models optimized for each task. The hyperparameter tuning unit, for example, uses the generation AI to automatically adjust hyperparameters that are optimal for different tasks. For example, different learning rates and batch sizes are set for image recognition tasks and natural language processing tasks. The generation AI, for example, sets an optimal learning rate for image recognition tasks and an optimal batch size for natural language processing tasks. The generation AI also sets optimal hyperparameters for different tasks, for example, to maximize model performance. In this way, by automatically adjusting hyperparameters that are optimal for different tasks, it is possible to provide models optimized for each task.

[0045] The hyperparameter tuning unit can use the generation AI to try multiple hyperparameter settings and develop a search algorithm to find the optimal settings. The hyperparameter tuning unit, for example, uses the generation AI to try multiple hyperparameter settings and develop a search algorithm to find the optimal settings. For example, the unit finds the optimal hyperparameters using grid search or random search. The generation AI, for example, tries different hyperparameter settings and evaluates their effects. The generation AI also develops, for example, a search algorithm to find the optimal hyperparameters and efficiently finds the optimal settings. In this way, by trying multiple hyperparameter settings and developing a search algorithm to find the optimal settings, the optimal hyperparameters can be found efficiently.

[0046] The hyperparameter tuning unit uses the generation AI to automatically adjust hyperparameters corresponding to image and audio data, and can build a multimodal model. The hyperparameter tuning unit, for example, uses the generation AI to automatically adjust optimal hyperparameters corresponding to image data. For example, it sets an optimal learning rate and batch size for an image classification task. The generation AI, for example, analyzes the features of the image data and sets optimal hyperparameters based on the patterns. The generation AI also automatically adjusts optimal hyperparameters corresponding to audio data, for example, and finds optimal settings for a speech recognition task. This makes it possible to build a multimodal model by automatically adjusting hyperparameters corresponding to image and audio data.

[0047] The hyperparameter tuning unit can use a generation AI to adjust hyperparameters in real time and provide a dynamic model. The hyperparameter tuning unit, for example, uses a generation AI to adjust hyperparameters in real time and provide a dynamic model. For example, it instantly sets optimal hyperparameters in response to user input. The generation AI, for example, analyzes data in real time and adjusts hyperparameters based on the results. The generation AI also, for example, adjusts model parameters in real time and provides a dynamic model. In this way, a dynamic model can be provided by adjusting hyperparameters in real time.

[0048] The automatic evaluation unit can use the generative AI to perform a comprehensive evaluation that combines different evaluation indicators to evaluate the overall performance of the model. The automatic evaluation unit, for example, uses the generative AI to perform a comprehensive evaluation that combines different evaluation indicators to evaluate the overall performance of the model. For example, the evaluation is performed by combining precision, recall, and F1 score. The generative AI evaluates the performance of the model based on, for example, different evaluation indicators and reports the results. The generative AI also suggests improvements to the model based on, for example, the results of the comprehensive evaluation. In this way, the overall performance of the model can be evaluated by performing a comprehensive evaluation that combines different evaluation indicators.

[0049] The automatic evaluation unit can use the generation AI to visualize the evaluation results, allowing the user to intuitively understand them. The automatic evaluation unit can, for example, use the generation AI to visualize the evaluation results, allowing the user to intuitively understand them. For example, the evaluation indicators can be displayed in graphs or charts. The generation AI can, for example, visually display the evaluation results, providing them in a format that is easy for the user to understand. The generation AI can also, for example, display the evaluation results in a dashboard format, allowing the user to check the evaluation results in real time. In this way, by visualizing the evaluation results, the user can intuitively understand them.

[0050] The automatic evaluation unit can use the generative AI to evaluate image and audio data and evaluate the performance of the multimodal model. The automatic evaluation unit, for example, uses the generative AI to evaluate image data and evaluate the performance of the multimodal model. For example, it evaluates the accuracy of the model for an image classification task. For example, the generative AI analyzes the features of the image data and evaluates the performance of the model based on the patterns. The generative AI also evaluates audio data, for example, and evaluates the performance of the model for a speech recognition task. In this way, the performance of the multimodal model can be evaluated by evaluating image and audio data.

[0051] The automatic evaluation unit can use a generative AI to perform evaluation in real time and evaluate the performance of a dynamic model. The automatic evaluation unit, for example, uses a generative AI to perform evaluation in real time and evaluate the performance of a dynamic model. For example, the automatic evaluation unit instantly evaluates the performance of a model in response to user input. For example, the generative AI analyzes data in real time and evaluates the performance of the model based on the results. In addition, the generative AI reports the evaluation results in real time and suggests improvements to the model. This makes it possible to evaluate the performance of a dynamic model by performing evaluation in real time.

[0052] The fine-tuning unit can use the generation AI to make fine adjustments for a specific task and provide a model optimized for each task. The fine-tuning unit, for example, uses the generation AI to make fine adjustments for a specific task and provide a model optimized for each task. For example, the fine-tuning unit adjusts model parameters for an image recognition task. For example, the generation AI sets optimal parameters for a specific task and maximizes model performance. The generation AI also sets optimal parameters for a natural language processing task, for example, and improves model performance. In this way, by making fine adjustments for a specific task, it is possible to provide a model optimized for each task.

[0053] The fine-tuning unit can use the generative AI to make fine adjustments corresponding to image and audio data and build a multimodal model. The fine-tuning unit, for example, uses the generative AI to make optimal fine adjustments corresponding to image data and build a multimodal model. For example, it adjusts the model parameters for an image classification task. For example, the generative AI analyzes the features of the image data and sets optimal parameters based on the patterns. The generative AI also makes optimal fine adjustments corresponding to audio data, for example, and finds optimal settings for a speech recognition task. In this way, a multimodal model can be built by making fine adjustments corresponding to image and audio data.

[0054] The fine-tuning unit can use the generation AI to perform fine-tuning in real time and provide a dynamic model. The fine-tuning unit can use, for example, the generation AI to perform fine-tuning in real time and provide a dynamic model. For example, the fine-tuning unit can instantly adjust model parameters in response to user input. For example, the generation AI can analyze data in real time and adjust parameters based on the results. The generation AI can also adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by performing fine-tuning in real time.

[0055] The model compression unit can use a generative AI to perform hybrid compression that combines different compression methods and provide an optimal compression model. The model compression unit, for example, uses a generative AI to perform hybrid compression that combines different compression methods and provide an optimal compression model. For example, it proposes a compression method that combines pruning and quantization. The generative AI, for example, combines different compression methods and evaluates their effectiveness. The generative AI also develops, for example, a search algorithm to find the optimal compression method and efficiently provides an optimal compression model. This makes it possible to provide an optimal compression model by performing hybrid compression that combines different compression methods.

[0056] The model compression unit can use the generation AI to evaluate the effect of compression and develop a search algorithm for finding the optimal compression method. The model compression unit, for example, uses the generation AI to evaluate the effect of compression and develop a search algorithm for finding the optimal compression method. For example, different compression methods are tried and their effects are compared. The generation AI, for example, tries different compression methods and evaluates their effects. The generation AI also, for example, develops a search algorithm for finding the optimal compression method and efficiently finds the optimal method. In this way, by evaluating the effect of compression and developing a search algorithm for finding the optimal compression method, the optimal compression method can be efficiently found.

[0057] The model compression unit uses a generative AI to perform model compression corresponding to image and audio data, and can build a multimodal lightweight model. The model compression unit, for example, uses a generative AI to perform optimal model compression corresponding to image data and build a multimodal lightweight model. For example, it reduces model parameters for an image classification task. For example, the generative AI analyzes the features of the image data and applies an optimal compression method based on the patterns. In addition, the generative AI performs optimal model compression corresponding to audio data, and finds optimal settings for a speech recognition task. In this way, a multimodal lightweight model can be built by performing model compression corresponding to image and audio data.

[0058] The model compression unit can perform model compression in real time using a generative AI to provide a dynamic lightweight model. The model compression unit can perform model compression in real time using, for example, a generative AI to provide a dynamic lightweight model. For example, the model compression unit can instantly reduce model parameters in response to user input. For example, the generative AI can analyze data in real time and reduce model parameters based on the results. The generative AI can also adjust model parameters in real time to provide a dynamic lightweight model. In this way, a dynamic lightweight model can be provided by performing model compression in real time.

[0059] The continuous learning unit can use the generative AI to develop a continuous learning method for improving adaptability to new data. The continuous learning unit, for example, uses the generative AI to develop a continuous learning method for improving adaptability to new data. For example, the continuous learning unit updates the model using an online learning algorithm. The generative AI, for example, incorporates new data and updates the model based on that data. The generative AI also, for example, collects data from different data sources and updates the model based on that data. In this way, the adaptability of the model can be improved by developing a continuous learning method for improving adaptability to new data.

[0060] The continuous learning unit can use the generation AI to evaluate the effectiveness of continuous learning and develop a search algorithm for finding the optimal learning method. The continuous learning unit, for example, uses the generation AI to evaluate the effectiveness of continuous learning and develop a search algorithm for finding the optimal learning method. For example, different continuous learning methods are tried and their effects are compared. The generation AI, for example, tries different continuous learning methods and evaluates their effects. The generation AI also, for example, develops a search algorithm for finding the optimal learning method and efficiently finds the optimal method. In this way, by evaluating the effectiveness of continuous learning and developing a search algorithm for finding the optimal learning method, it is possible to efficiently find the optimal learning method.

[0061] The continuous learning unit can use the generative AI to perform continuous learning in real time and provide a dynamic model. The continuous learning unit can use, for example, the generative AI to perform continuous learning in real time and provide a dynamic model. For example, the continuous learning unit can instantly update model parameters in response to user input. For example, the generative AI can analyze data in real time and update model parameters based on the results. The generative AI can also adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by performing continuous learning in real time.

[0062] The system according to the embodiment is not limited to the above-described example, and various modifications are possible, for example, as follows.

[0063] The integrated generative AI system may further include a user feedback collection unit. The user feedback collection unit collects feedback from users and suggests improvements to the model based on the feedback. For example, the user may evaluate the generated text and adjust the model parameters based on the evaluation results. The user feedback collection unit may also monitor user usage and optimize the model based on usage frequency and patterns. This allows the model's performance to be continuously improved by utilizing user feedback.

[0064] The data augmentation unit uses generative AI to generate data corresponding to different time periods and seasons, building a dataset with temporal diversity. For example, it can generate text data specific to morning hours or summer. The generative AI can learn from existing data corresponding to time periods and seasons, for example, and generate new data based on those patterns. It can also generate data related to specific events or holidays. This allows for the construction of a dataset with temporal diversity, improving the versatility of the model.

[0065] The data augmentation unit can use generative AI to generate data specialized for specific user groups and build personalized datasets. For example, it can generate text data specialized for specific age groups or occupations. The generative AI can, for example, learn existing data for specific user groups and generate new data based on those patterns. It can also generate data based on specific hobbies and interests. This allows for the creation of personalized models by building datasets specialized for specific user groups.

[0066] The data cleaning unit can use the generation AI to detect and automatically delete duplicate data. For example, if there are multiple pieces of text data with the same content, the duplicates are detected and deleted. The generation AI can, for example, analyze the content of the text data and identify the duplicated parts. It can also select and retain the most reliable data from among the duplicated data. This makes it possible to improve the efficiency and reliability of the dataset by removing duplicate data.

[0067] The data cleaning unit can use the generative AI to check the consistency of the data and automatically correct inconsistent data. For example, it corrects data when different information is stated within the same context. The generative AI, for example, understands the context, detects inconsistent information, and corrects it to appropriate information. The generative AI can also, for example, remove duplicate information and build a consistent dataset. This can improve the reliability of the dataset by maintaining the consistency of the data.

[0068] The data cleaning unit uses the generation AI to evaluate the reliability of the data and automatically remove unreliable data. For example, it deletes information whose source is unclear. The generation AI, for example, detects unreliable information and removes that part from the dataset. The generation AI also, for example, selects only highly reliable information and builds a dataset. This makes it possible to improve the quality of the dataset by removing unreliable data.

[0069] The processing flow of the first embodiment will be briefly explained below.

[0070] Step 1: The data augmentation unit uses generative AI to generate new data based on an existing dataset and expand the dataset. For example, the generative AI generates new sentences based on existing sentences, increasing the diversity of the dataset. The input to the generative AI is a prompt containing instructions for data augmentation, and the generative AI generates new data based on the prompt. Step 2: The data cleaning part uses the generative AI to automatically detect and clean noisy or inaccurate data in the dataset. For example, the generative AI detects and corrects typos and grammatical errors. The input to the generative AI is a prompt containing instructions for data cleaning, and the generative AI cleans the data based on the prompt. Step 3: The architecture generation unit uses the generative AI to automatically generate the optimal model architecture. For example, the generative AI designs the optimal neural network structure for a given task and proposes that architecture. The input to the generative AI is a prompt containing instructions for generating the architecture, and the generative AI generates the architecture based on the prompt. Step 4: The training data optimization unit proposes methods to optimize the training data using the generative AI. For example, the generative AI adjusts the balance of the data and builds an optimal dataset to improve the model's performance. The input to the generative AI is a prompt containing instructions for optimizing the training data, and the generative AI optimizes the data based on the prompt. Step 5: The hyperparameter tuning part uses the generative AI to automatically tune the model's hyperparameters to find optimal settings. For example, the generative AI adjusts hyperparameters such as the learning rate and batch size to maximize model performance. The generative AI receives inputs from prompts containing instructions for tuning the hyperparameters, and the generative AI adjusts the hyperparameters based on the prompts. Step 6: The automatic evaluation unit uses the generative AI to automatically evaluate the model's performance. For example, the generative AI measures the model's precision and recall using test data and reports the results. The generative AI receives input from a prompt containing instructions for model evaluation, and the generative AI evaluates the model based on the prompt. Step 7: The fine-tuner uses the generative AI to fine-tune the model to improve its performance. For example, the generative AI performs fine-tuning to improve the model's adaptability to a specific task. The input to the generative AI is a prompt containing fine-tuning instructions, and the generative AI fine-tunes the model based on the prompt. Step 8: The model compression unit uses the generative AI to compress the size of the model, enabling efficient execution. For example, the generative AI may reduce unnecessary parameters to make the model lighter. The input to the generative AI is a prompt containing instructions for model compression, and the generative AI compresses the model based on the prompt. Step 9: The continuous learning component uses the generative AI to help the model continuously learn from new data. For example, the generative AI can incorporate new data and suggest ways to update the model. The input to the generative AI is a prompt containing instructions for continuous learning, and the generative AI continuously learns based on the prompt.

[0071] (Example 2) The integrated generative AI system according to an embodiment of the present invention is a system that streamlines the process of creating large-scale language models and maximizes model performance. As a result, the integrated generative AI system utilizes generative AI at each stage, from data collection to model training and optimization, to create efficient, high-performance large-scale language models.

[0072] The integrated generative AI system according to the embodiment includes a data augmentation unit, a data cleaning unit, an architecture generation unit, a training data optimization unit, a hyperparameter tuning unit, an automatic evaluation unit, a fine-tuning unit, a model compression unit, and a continuous learning unit. The data augmentation unit uses a generative AI to generate new data based on an existing dataset and expand the dataset. For example, the generative AI generates new sentences based on existing sentences to increase the diversity of the dataset. The generative AI receives inputs from prompts containing instructions for data augmentation, and the generative AI generates new data based on the prompts. The data cleaning unit uses a generative AI to automatically detect and clean noisy or inaccurate data in a dataset. For example, the generative AI detects and corrects typos and grammatical errors. The generative AI receives inputs from prompts containing instructions for data cleaning, and the generative AI cleans the data based on the prompts. The architecture generation unit uses a generative AI to automatically generate an optimal model architecture. For example, the generative AI designs the structure of a neural network optimal for a given task and proposes its architecture. The generative AI receives inputs from prompts containing instructions for architecture generation, and the generative AI generates an architecture based on the prompts. The training data optimization unit uses the generative AI to propose methods for optimizing training data. For example, the generative AI adjusts the balance of data to build an optimal dataset to improve model performance. The generative AI receives inputs from prompts containing instructions for optimizing training data, and the generative AI optimizes the data based on those prompts. The hyperparameter tuning unit uses the generative AI to automatically adjust the model's hyperparameters to find optimal settings. For example, the generative AI adjusts hyperparameters such as the learning rate and batch size to maximize model performance. The generative AI receives inputs from prompts containing instructions for tuning the hyperparameters, and the generative AI adjusts the hyperparameters based on those prompts. The automatic evaluation unit uses the generative AI to automatically evaluate the model's performance. For example, the generative AI uses test data to measure the model's precision and recall and reports the results.The input to the generative AI is a prompt containing instructions for model evaluation, and the generative AI evaluates the model based on the prompt. The fine-tuning unit uses the generative AI to fine-tune the model to improve its performance. For example, the generative AI performs fine-tuning to increase the adaptability of the model to a specific task. The input to the generative AI is a prompt containing instructions for fine-tuning, and the generative AI fine-tunes the model based on the prompt. The model compression unit uses the generative AI to compress the model size to enable efficient execution. For example, the generative AI reduces unnecessary parameters to make the model lighter. The input to the generative AI is a prompt containing instructions for model compression, and the generative AI compresses the model based on the prompt. The continuous learning unit uses the generative AI to help the model continuously learn from new data. For example, the generative AI incorporates new data and proposes a method for updating the model. The input to the generative AI is a prompt containing instructions for continuous learning, and the generative AI performs continuous learning based on the prompt. As a result, the integrated generative AI system according to the embodiment can streamline the process of creating large-scale language models and maximize model performance. For example, by building a high-quality dataset through data augmentation and cleaning, and training the model with optimal architecture and hyperparameters, it is possible to create a highly accurate language model. Furthermore, model compression and continuous learning enable efficient use in production environments.

[0073] The data augmentation unit can use the generative AI to generate text data with specific emotions and build an emotion-based dataset. The data augmentation unit, for example, uses the generative AI to generate text data with specific emotions, such as joy or sadness. For example, it uses a sentiment analysis model to extract an emotion score from existing text data and generate new text based on that score. For example, to generate sentences with positive emotions, the generative AI learns from existing data with positive emotions and generates new sentences based on those patterns. Also, to generate sentences with negative emotions, the generative AI learns from existing data with negative emotions and generates new sentences based on those patterns. In this way, by building an emotion-based dataset, the accuracy of the sentiment analysis model can be improved.

[0074] The data augmentation unit can use the generative AI to generate data from different cultures and languages ​​and create a global dataset. The data augmentation unit, for example, uses the generative AI to generate text data from different cultures and languages. For example, it generates multilingual data such as English, Japanese, and French to build a global dataset. For example, to generate sentences from different cultures, the generative AI learns existing data from different cultures and generates new sentences based on their patterns. Also, for example, to generate sentences in different languages, the generative AI learns existing data from different languages ​​and generates new sentences based on their patterns. This allows the versatility of the model to be improved by creating a global dataset that includes data from different cultures and languages.

[0075] The data augmentation unit uses generative AI to generate data specialized for a specific industry or field of expertise, thereby building a highly specialized dataset. The data augmentation unit, for example, uses generative AI to generate text data specialized for a specific industry, such as medicine or law. For example, new text is generated based on medical papers or legal documents. The generative AI, for example, learns medical terminology and writing style and generates new medical-related text based on those patterns. The generative AI also learns legal terminology and writing style and generates new legal-related text based on those patterns. This makes it possible to create a highly specialized model by building a dataset specialized for a specific industry or field of expertise.

[0076] The data cleaning unit can use the generative AI to perform sentiment analysis and automatically correct data with negative sentiment. For example, the data cleaning unit uses the generative AI to perform sentiment analysis of text data and correct parts with negative sentiment. For example, it replaces negative expressions with positive expressions. For example, the generative AI detects sentences with negative sentiment and converts those parts into sentences with positive sentiment. The generative AI also replaces words or phrases with negative sentiment with positive ones. In this way, the quality of the dataset can be improved by correcting data with negative sentiment.

[0077] The data cleaning unit can use the generation AI to check the consistency of the data and automatically correct inconsistent data. The data cleaning unit, for example, uses the generation AI to check the consistency of text data and automatically correct inconsistent parts. For example, it corrects when different information is stated within the same context. The generation AI, for example, understands the context, detects inconsistent information, and corrects it to appropriate information. The generation AI also, for example, removes duplicate information and builds a consistent dataset. This makes it possible to maintain the consistency of the data and improve the reliability of the dataset.

[0078] The data cleaning unit can use the generation AI to evaluate the reliability of data and automatically remove low-reliability data. The data cleaning unit, for example, uses the generation AI to evaluate the reliability of text data and automatically remove low-reliability parts. For example, it deletes information whose source is unclear. The generation AI, for example, detects low-reliability information and removes that part from the dataset. The generation AI also, for example, selects only high-reliability information and builds a dataset. This makes it possible to improve the quality of the dataset by removing low-reliability data.

[0079] The data cleaning unit can use the generation AI to clean image and audio data and prepare a multimodal dataset. The data cleaning unit, for example, uses the generation AI to remove noise and improve the resolution of image data and clean it. For example, it can make blurry images clear. The generation AI, for example, evaluates the quality of image data and removes noise. The generation AI can also convert low-resolution images to high resolution. The generation AI can also clean audio data by removing noise and improving sound quality, for example. For example, it can remove background noise and make audio clear. In this way, cleaning image and audio data can improve the quality of the multimodal dataset.

[0080] The data cleaning unit can use the generation AI to perform data cleaning in real time and maintain a dynamic dataset. The data cleaning unit can, for example, use the generation AI to clean text data in real time and maintain a dynamic dataset. For example, it can instantly correct typos in response to user input. For example, the generation AI can check the consistency of the data in real time and correct contradictory information. In addition, the generation AI can, for example, evaluate the reliability of the data in real time and remove unreliable data. In this way, the quality of the dynamic dataset can be maintained by performing data cleaning in real time.

[0081] The architecture generation unit can use generative AI to automatically generate architectures that are optimal for different tasks and provide models optimized for each task. The architecture generation unit, for example, uses generative AI to automatically generate model architectures that are optimal for different tasks. For example, different neural network structures are proposed for image recognition tasks and natural language processing tasks. For example, the generative AI proposes a CNN architecture for image recognition tasks and an RNN architecture for natural language processing tasks. The generative AI also sets optimal hyperparameters for different tasks, for example, to maximize model performance. In this way, by automatically generating architectures that are optimal for different tasks, it is possible to provide models optimized for each task.

[0082] The architecture generation unit can use generative AI to generate a hybrid model that combines multiple architectures. The architecture generation unit, for example, uses generative AI to generate a hybrid model that combines different architectures. For example, it proposes a model that combines CNN and RNN. The generative AI generates a model that combines, for example, the feature extraction capabilities of CNN and the time series data processing capabilities of RNN. The generative AI also proposes a hybrid model that utilizes the strengths of different architectures, for example, to improve the performance of the model. In this way, by generating a hybrid model that combines multiple architectures, it is possible to provide a model that utilizes the strengths of each architecture.

[0083] The architecture generation unit uses generative AI to automatically generate architectures that correspond to image and audio data, and can build multimodal models. The architecture generation unit, for example, uses generative AI to automatically generate an optimal architecture that corresponds to image data. For example, it proposes a CNN architecture specialized for image classification tasks. The generative AI, for example, learns the features of the image data and generates an optimal architecture based on those patterns. The generative AI also automatically generates an optimal architecture that corresponds to audio data. For example, it proposes an RNN architecture specialized for audio recognition tasks. This makes it possible to build a multimodal model by automatically generating an architecture that corresponds to image and audio data.

[0084] The architecture generation unit can use the generative AI to estimate the user's emotions and adjust the architecture based on those emotions. The architecture generation unit, for example, uses the generative AI to estimate the user's emotions and adjust the model architecture based on those emotions. For example, it proposes an architecture that is optimal for data containing positive emotions. The generative AI, for example, analyzes the user's emotions and adjusts the architecture parameters based on those emotions. The generative AI can also propose an architecture that is optimal for data containing negative emotions, for example, and improve the performance of the model. In this way, by adjusting the architecture based on the user's emotions, it is possible to provide a model that is adapted to emotions.

[0085] The architecture generation unit can use a generative AI to generate an architecture in real time and provide a dynamic model. The architecture generation unit can use, for example, a generative AI to generate a model architecture in real time and provide a dynamic model. For example, the architecture generation unit can instantly propose an optimal architecture in response to user input. For example, the generative AI can analyze data in real time and generate an architecture based on the results. Furthermore, the generative AI can adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by generating an architecture in real time.

[0086] The training data optimization unit can use the generative AI to optimize training data based on emotions and train a model that adapts to emotions. The training data optimization unit, for example, uses the generative AI to optimize training data based on emotions. For example, data with positive emotions is preferentially selected to train a model. The generative AI, for example, selects data based on emotion scores and builds an optimal training dataset. The generative AI also, for example, excludes data with negative emotions to improve model performance. In this way, by optimizing the training data based on emotions, a model that adapts to emotions can be trained.

[0087] The training data optimization unit can use the generation AI to adjust the balance of the data and build a dataset that is not biased towards a particular class. The training data optimization unit, for example, uses the generation AI to adjust the balance of the training data and build a dataset that is not biased towards a particular class. For example, it equalizes the amount of data for each class. The generation AI, for example, analyzes the distribution of the data and builds a balanced dataset. In addition, the generation AI, for example, avoids data duplication in order to maintain data balance. In this way, by adjusting the balance of the data, it is possible to build a dataset that is not biased towards a particular class.

[0088] The training data optimization unit can use the generative AI to collect and integrate data from different data sources to increase data diversity. The training data optimization unit, for example, uses the generative AI to collect and integrate data from different data sources to increase data diversity. For example, it collects news articles and blog articles. The generative AI, for example, collects data from different data sources and integrates the data to build a diverse dataset. The generative AI also, for example, integrates data in different formats to increase the diversity of the dataset. This makes it possible to increase data diversity by collecting and integrating data from different data sources.

[0089] The training data optimization unit can use the generative AI to optimize training data for image and audio data and train a multimodal model. The training data optimization unit, for example, uses the generative AI to optimize training data for image data and train a multimodal model. For example, it builds a dataset that is optimal for an image classification task. For example, the generative AI analyzes the features of the image data and builds an optimal training dataset based on the patterns. Furthermore, the generative AI can optimize training data for audio data and build an optimal dataset for a speech recognition task. In this way, it is possible to train a multimodal model by optimizing the training data for image and audio data.

[0090] The training data optimization unit can use the generative AI to estimate the user's emotions and optimize the training data based on those emotions. The training data optimization unit, for example, uses the generative AI to estimate the user's emotions and optimize the training data based on those emotions. For example, it preferentially selects data with positive emotions. For example, the generative AI analyzes the user's emotions, selects data based on those emotions, and builds an optimal training dataset. The generative AI also excludes data with negative emotions, for example, to improve the model's performance. In this way, by optimizing the training data based on the user's emotions, it is possible to train a model that adapts to emotions.

[0091] The training data optimization unit can use the generative AI to optimize the training data in real time and train a dynamic model. The training data optimization unit can use, for example, the generative AI to optimize the training data in real time and train a dynamic model. For example, the generative AI can instantly update the dataset in response to user input. For example, the generative AI can analyze data in real time and optimize the training data based on the results. In addition, the generative AI can adjust the balance of the data in real time and build a dataset that is not biased toward a particular class. This makes it possible to train a dynamic model by optimizing the training data in real time.

[0092] The hyperparameter tuning unit can use the generation AI to automatically adjust hyperparameters that are optimal for different tasks and provide models optimized for each task. The hyperparameter tuning unit, for example, uses the generation AI to automatically adjust hyperparameters that are optimal for different tasks. For example, different learning rates and batch sizes are set for image recognition tasks and natural language processing tasks. The generation AI, for example, sets an optimal learning rate for image recognition tasks and an optimal batch size for natural language processing tasks. The generation AI also sets optimal hyperparameters for different tasks, for example, to maximize model performance. In this way, by automatically adjusting hyperparameters that are optimal for different tasks, it is possible to provide models optimized for each task.

[0093] The hyperparameter tuning unit can use the generation AI to try multiple hyperparameter settings and develop a search algorithm to find the optimal settings. The hyperparameter tuning unit, for example, uses the generation AI to try multiple hyperparameter settings and develop a search algorithm to find the optimal settings. For example, the unit finds the optimal hyperparameters using grid search or random search. The generation AI, for example, tries different hyperparameter settings and evaluates their effects. The generation AI also develops, for example, a search algorithm to find the optimal hyperparameters and efficiently finds the optimal settings. In this way, by trying multiple hyperparameter settings and developing a search algorithm to find the optimal settings, the optimal hyperparameters can be found efficiently.

[0094] The hyperparameter tuning unit uses the generation AI to automatically adjust hyperparameters corresponding to image and audio data, and can build a multimodal model. The hyperparameter tuning unit, for example, uses the generation AI to automatically adjust optimal hyperparameters corresponding to image data. For example, it sets an optimal learning rate and batch size for an image classification task. The generation AI, for example, analyzes the features of the image data and sets optimal hyperparameters based on the patterns. The generation AI also automatically adjusts optimal hyperparameters corresponding to audio data, for example, and finds optimal settings for a speech recognition task. This makes it possible to build a multimodal model by automatically adjusting hyperparameters corresponding to image and audio data.

[0095] The hyperparameter tuning unit can use the generation AI to estimate the user's emotions and adjust hyperparameters based on those emotions. The hyperparameter tuning unit, for example, uses the generation AI to estimate the user's emotions and adjust hyperparameters based on those emotions. For example, it sets an optimal learning rate and batch size for data with positive emotions. The generation AI, for example, analyzes the user's emotions and adjusts hyperparameters based on those emotions. The generation AI also sets optimal hyperparameters for data with negative emotions, for example, to improve model performance. In this way, by adjusting hyperparameters based on the user's emotions, it is possible to build a model that adapts to emotions.

[0096] The hyperparameter tuning unit can use a generation AI to adjust hyperparameters in real time and provide a dynamic model. The hyperparameter tuning unit, for example, uses a generation AI to adjust hyperparameters in real time and provide a dynamic model. For example, it instantly sets optimal hyperparameters in response to user input. The generation AI, for example, analyzes data in real time and adjusts hyperparameters based on the results. The generation AI also, for example, adjusts model parameters in real time and provides a dynamic model. In this way, a dynamic model can be provided by adjusting hyperparameters in real time.

[0097] The automatic evaluation unit can use the generative AI to perform an evaluation based on emotions and evaluate the performance of the emotion-adaptive model. The automatic evaluation unit, for example, uses the generative AI to perform an evaluation based on emotions and evaluate the performance of the emotion-adaptive model. For example, it evaluates the accuracy of the model for data with positive emotions. The generative AI evaluates the performance of the model based on, for example, the emotion score and reports the results. The generative AI also evaluates the performance of the model for data with negative emotions, for example, and suggests areas for improvement in the model. In this way, the performance of the emotion-adaptive model can be evaluated by performing an evaluation based on emotions.

[0098] The automatic evaluation unit can use the generative AI to perform a comprehensive evaluation that combines different evaluation indicators to evaluate the overall performance of the model. The automatic evaluation unit, for example, uses the generative AI to perform a comprehensive evaluation that combines different evaluation indicators to evaluate the overall performance of the model. For example, the evaluation is performed by combining precision, recall, and F1 score. The generative AI evaluates the performance of the model based on, for example, different evaluation indicators and reports the results. The generative AI also suggests improvements to the model based on, for example, the results of the comprehensive evaluation. In this way, the overall performance of the model can be evaluated by performing a comprehensive evaluation that combines different evaluation indicators.

[0099] The automatic evaluation unit can use the generation AI to visualize the evaluation results, allowing the user to intuitively understand them. The automatic evaluation unit can, for example, use the generation AI to visualize the evaluation results, allowing the user to intuitively understand them. For example, the evaluation indicators can be displayed in graphs or charts. The generation AI can, for example, visually display the evaluation results, providing them in a format that is easy for the user to understand. The generation AI can also, for example, display the evaluation results in a dashboard format, allowing the user to check the evaluation results in real time. In this way, by visualizing the evaluation results, the user can intuitively understand them.

[0100] The automatic evaluation unit can use the generative AI to evaluate image and audio data and evaluate the performance of the multimodal model. The automatic evaluation unit, for example, uses the generative AI to evaluate image data and evaluate the performance of the multimodal model. For example, it evaluates the accuracy of the model for an image classification task. For example, the generative AI analyzes the features of the image data and evaluates the performance of the model based on the patterns. The generative AI also evaluates audio data, for example, and evaluates the performance of the model for a speech recognition task. In this way, the performance of the multimodal model can be evaluated by evaluating image and audio data.

[0101] The automatic evaluation unit can use the generation AI to estimate the user's emotions and adjust the evaluation results based on those emotions. The automatic evaluation unit, for example, uses the generation AI to estimate the user's emotions and adjust the evaluation results based on those emotions. For example, the evaluation results are adjusted for data with positive emotions. The generation AI, for example, analyzes the user's emotions and corrects the evaluation results based on those emotions. The generation AI also adjusts the evaluation results for data with negative emotions, for example, to improve the performance of the model. In this way, by adjusting the evaluation results based on the user's emotions, an evaluation adapted to emotions can be performed.

[0102] The automatic evaluation unit can use a generative AI to perform evaluation in real time and evaluate the performance of a dynamic model. The automatic evaluation unit, for example, uses a generative AI to perform evaluation in real time and evaluate the performance of a dynamic model. For example, the automatic evaluation unit instantly evaluates the performance of a model in response to user input. For example, the generative AI analyzes data in real time and evaluates the performance of the model based on the results. In addition, the generative AI reports the evaluation results in real time and suggests improvements to the model. This makes it possible to evaluate the performance of a dynamic model by performing evaluation in real time.

[0103] The fine-tuning unit can use the generative AI to perform emotion-based fine-tuning and build a model that adapts to emotions. The fine-tuning unit, for example, uses the generative AI to perform emotion-based fine-tuning and build a model that adapts to emotions. For example, it adjusts model parameters for data that has positive emotions. The generative AI, for example, adjusts model parameters based on emotion scores and finds optimal settings. The generative AI also adjusts model parameters for data that has negative emotions, for example, to improve model performance. In this way, it is possible to build a model that adapts to emotions by performing emotion-based fine-tuning.

[0104] The fine-tuning unit can use the generation AI to make fine adjustments for a specific task and provide a model optimized for each task. The fine-tuning unit, for example, uses the generation AI to make fine adjustments for a specific task and provide a model optimized for each task. For example, the fine-tuning unit adjusts model parameters for an image recognition task. For example, the generation AI sets optimal parameters for a specific task and maximizes model performance. The generation AI also sets optimal parameters for a natural language processing task, for example, and improves model performance. In this way, by making fine adjustments for a specific task, it is possible to provide a model optimized for each task.

[0105] The fine-tuning unit can use the generative AI to make fine adjustments corresponding to image and audio data and build a multimodal model. The fine-tuning unit, for example, uses the generative AI to make optimal fine adjustments corresponding to image data and build a multimodal model. For example, it adjusts the model parameters for an image classification task. For example, the generative AI analyzes the features of the image data and sets optimal parameters based on the patterns. The generative AI also makes optimal fine adjustments corresponding to audio data, for example, and finds optimal settings for a speech recognition task. In this way, a multimodal model can be built by making fine adjustments corresponding to image and audio data.

[0106] The fine-tuning unit can use the generation AI to estimate the user's emotions and perform fine-tuning based on those emotions. The fine-tuning unit, for example, uses the generation AI to estimate the user's emotions and perform fine-tuning based on those emotions. For example, the fine-tuning unit adjusts the model parameters for data containing positive emotions. For example, the generation AI analyzes the user's emotions and adjusts the parameters based on those emotions. The generation AI also sets optimal parameters for data containing negative emotions, for example, to improve the model's performance. In this way, by performing fine-tuning based on the user's emotions, it is possible to build a model that adapts to emotions.

[0107] The fine-tuning unit can use the generation AI to perform fine-tuning in real time and provide a dynamic model. The fine-tuning unit can use, for example, the generation AI to perform fine-tuning in real time and provide a dynamic model. For example, the fine-tuning unit can instantly adjust model parameters in response to user input. For example, the generation AI can analyze data in real time and adjust parameters based on the results. The generation AI can also adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by performing fine-tuning in real time.

[0108] The model compression unit can use generative AI to perform model compression based on emotions and build a lightweight model that adapts to emotions. The model compression unit, for example, uses generative AI to perform model compression based on emotions and build a lightweight model that adapts to emotions. For example, it reduces model parameters for data with positive emotions. For example, the generative AI reduces model parameters based on emotion scores and finds optimal settings. The generative AI can also reduce model parameters for data with negative emotions, for example, to improve model performance. In this way, it is possible to build a lightweight model that adapts to emotions by performing model compression based on emotions.

[0109] The model compression unit can use a generative AI to perform hybrid compression that combines different compression methods and provide an optimal compression model. The model compression unit, for example, uses a generative AI to perform hybrid compression that combines different compression methods and provide an optimal compression model. For example, it proposes a compression method that combines pruning and quantization. The generative AI, for example, combines different compression methods and evaluates their effectiveness. The generative AI also develops, for example, a search algorithm to find the optimal compression method and efficiently provides an optimal compression model. This makes it possible to provide an optimal compression model by performing hybrid compression that combines different compression methods.

[0110] The model compression unit can use the generation AI to evaluate the effect of compression and develop a search algorithm for finding the optimal compression method. The model compression unit, for example, uses the generation AI to evaluate the effect of compression and develop a search algorithm for finding the optimal compression method. For example, different compression methods are tried and their effects are compared. The generation AI, for example, tries different compression methods and evaluates their effects. The generation AI also, for example, develops a search algorithm for finding the optimal compression method and efficiently finds the optimal method. In this way, by evaluating the effect of compression and developing a search algorithm for finding the optimal compression method, the optimal compression method can be efficiently found.

[0111] The model compression unit uses a generative AI to perform model compression corresponding to image and audio data, and can build a multimodal lightweight model. The model compression unit, for example, uses a generative AI to perform optimal model compression corresponding to image data and build a multimodal lightweight model. For example, it reduces model parameters for an image classification task. For example, the generative AI analyzes the features of the image data and applies an optimal compression method based on the patterns. In addition, the generative AI performs optimal model compression corresponding to audio data, and finds optimal settings for a speech recognition task. In this way, a multimodal lightweight model can be built by performing model compression corresponding to image and audio data.

[0112] The model compression unit can use a generative AI to estimate a user's emotion and perform model compression based on that emotion. The model compression unit, for example, uses a generative AI to estimate a user's emotion and perform model compression based on that emotion. For example, it reduces model parameters for data containing positive emotions. For example, the generative AI analyzes the user's emotion and reduces model parameters based on that emotion. The generative AI also applies an optimal compression method to data containing negative emotions, for example, to improve model performance. In this way, by performing model compression based on the user's emotion, it is possible to build a lightweight model that adapts to emotions.

[0113] The model compression unit can perform model compression in real time using a generative AI to provide a dynamic lightweight model. The model compression unit can perform model compression in real time using, for example, a generative AI to provide a dynamic lightweight model. For example, the model compression unit can instantly reduce model parameters in response to user input. For example, the generative AI can analyze data in real time and reduce model parameters based on the results. The generative AI can also adjust model parameters in real time to provide a dynamic lightweight model. In this way, a dynamic lightweight model can be provided by performing model compression in real time.

[0114] The continuous learning unit can use the generative AI to perform continuous learning based on emotions and build a model that adapts to emotions. The continuous learning unit, for example, uses the generative AI to perform continuous learning based on emotions and build a model that adapts to emotions. For example, it updates the model parameters for data that has positive emotions. The generative AI, for example, updates the model parameters based on the emotion score and finds the optimal settings. The generative AI also updates the model parameters for data that has negative emotions, for example, to improve the model's performance. In this way, a model that adapts to emotions can be built by performing continuous learning based on emotions.

[0115] The continuous learning unit can use the generative AI to develop a continuous learning method for improving adaptability to new data. The continuous learning unit, for example, uses the generative AI to develop a continuous learning method for improving adaptability to new data. For example, the continuous learning unit updates the model using an online learning algorithm. The generative AI, for example, incorporates new data and updates the model based on that data. The generative AI also, for example, collects data from different data sources and updates the model based on that data. In this way, the adaptability of the model can be improved by developing a continuous learning method for improving adaptability to new data.

[0116] The continuous learning unit can use the generation AI to evaluate the effectiveness of continuous learning and develop a search algorithm for finding the optimal learning method. The continuous learning unit, for example, uses the generation AI to evaluate the effectiveness of continuous learning and develop a search algorithm for finding the optimal learning method. For example, different continuous learning methods are tried and their effects are compared. The generation AI, for example, tries different continuous learning methods and evaluates their effects. The generation AI also, for example, develops a search algorithm for finding the optimal learning method and efficiently finds the optimal method. In this way, by evaluating the effectiveness of continuous learning and developing a search algorithm for finding the optimal learning method, it is possible to efficiently find the optimal learning method.

[0117] The continuous learning unit can use the generation AI to estimate the user's emotions and perform continuous learning based on those emotions. The continuous learning unit, for example, uses the generation AI to estimate the user's emotions and perform continuous learning based on those emotions. For example, the continuous learning unit continuously updates the model parameters for data containing positive emotions. For example, the generation AI analyzes the user's emotions and updates the model parameters based on those emotions. The generation AI also updates the model parameters for data containing negative emotions, for example, to improve the model's performance. In this way, by performing continuous learning based on the user's emotions, it is possible to build a model that adapts to emotions.

[0118] The continuous learning unit can use the generative AI to perform continuous learning in real time and provide a dynamic model. The continuous learning unit can use, for example, the generative AI to perform continuous learning in real time and provide a dynamic model. For example, the continuous learning unit can instantly update model parameters in response to user input. For example, the generative AI can analyze data in real time and update model parameters based on the results. The generative AI can also adjust model parameters in real time and provide a dynamic model. In this way, a dynamic model can be provided by performing continuous learning in real time.

[0119] The system according to the embodiment is not limited to the above-described example, and various modifications are possible, for example, as follows.

[0120] The integrated generative AI system may further include a user feedback collection unit. The user feedback collection unit collects feedback from users and suggests improvements to the model based on the feedback. For example, the user may evaluate the generated text and adjust the model parameters based on the evaluation results. The user feedback collection unit may also monitor user usage and optimize the model based on usage frequency and patterns. This allows the model's performance to be continuously improved by utilizing user feedback.

[0121] The data augmentation unit uses generative AI to estimate the user's emotions and generate new data based on those emotions. For example, if the user has positive emotions, it can generate positive text data based on those emotions. Conversely, if the user has negative emotions, it can generate negative text data based on those emotions. Furthermore, it is possible to monitor changes in the user's emotions in real time and adjust the content of the generated data in response to those changes. This makes it possible to build a dataset that adapts to the user's emotions.

[0122] The data augmentation unit uses generative AI to generate data corresponding to different time periods and seasons, building a dataset with temporal diversity. For example, it can generate text data specific to morning hours or summer. The generative AI can learn from existing data corresponding to time periods and seasons, for example, and generate new data based on those patterns. It can also generate data related to specific events or holidays. This allows for the construction of a dataset with temporal diversity, improving the versatility of the model.

[0123] The data augmentation unit can use generative AI to generate data specialized for specific user groups and build personalized datasets. For example, it can generate text data specialized for specific age groups or occupations. The generative AI can, for example, learn existing data for specific user groups and generate new data based on those patterns. It can also generate data based on specific hobbies and interests. This allows for the creation of personalized models by building datasets specialized for specific user groups.

[0124] The data cleaning unit uses generative AI to perform sentiment analysis and prioritize data with positive sentiment. For example, it can detect text data with positive sentiment and retain that data in the dataset. Generative AI can also detect words and phrases with positive sentiment and highlight those parts. It can also remove data with negative sentiment. This allows the quality of the dataset to be improved by prioritizing data with positive sentiment.

[0125] The data cleaning unit can use the generation AI to detect and automatically delete duplicate data. For example, if there are multiple pieces of text data with the same content, the duplicates are detected and deleted. The generation AI can, for example, analyze the content of the text data and identify the duplicated parts. It can also select and retain the most reliable data from among the duplicated data. This makes it possible to improve the efficiency and reliability of the dataset by removing duplicate data.

[0126] The data cleaning unit uses generative AI to perform sentiment analysis and highlight data with specific sentiments. For example, it highlights text data with positive sentiment and leaves that data in the dataset. Generative AI can, for example, detect words and phrases with positive sentiment and highlight those parts. It can also remove data with negative sentiment. This makes it possible to build a sentiment-based dataset by highlighting data with specific sentiments.

[0127] The data cleaning unit can use the generative AI to check the consistency of the data and automatically correct inconsistent data. For example, it corrects data when different information is stated within the same context. The generative AI, for example, understands the context, detects inconsistent information, and corrects it to appropriate information. The generative AI can also, for example, remove duplicate information and build a consistent dataset. This can improve the reliability of the dataset by maintaining the consistency of the data.

[0128] The data cleaning unit uses generative AI to perform sentiment analysis and highlight data with specific sentiments. For example, it highlights text data with positive sentiment and leaves that data in the dataset. Generative AI can, for example, detect words and phrases with positive sentiment and highlight those parts. It can also remove data with negative sentiment. This makes it possible to build a sentiment-based dataset by highlighting data with specific sentiments.

[0129] The data cleaning unit uses the generation AI to evaluate the reliability of the data and automatically remove unreliable data. For example, it deletes information whose source is unclear. The generation AI, for example, detects unreliable information and removes that part from the dataset. The generation AI also, for example, selects only highly reliable information and builds a dataset. This makes it possible to improve the quality of the dataset by removing unreliable data.

[0130] The processing flow of the second embodiment will be briefly explained below.

[0131] Step 1: The data augmentation unit uses generative AI to generate new data based on an existing dataset and expand the dataset. For example, the generative AI generates new sentences based on existing sentences, increasing the diversity of the dataset. The input to the generative AI is a prompt containing instructions for data augmentation, and the generative AI generates new data based on the prompt. Step 2: The data cleaning part uses the generative AI to automatically detect and clean noisy or inaccurate data in the dataset. For example, the generative AI detects and corrects typos and grammatical errors. The input to the generative AI is a prompt containing instructions for data cleaning, and the generative AI cleans the data based on the prompt. Step 3: The architecture generation unit uses the generative AI to automatically generate the optimal model architecture. For example, the generative AI designs the optimal neural network structure for a given task and proposes that architecture. The input to the generative AI is a prompt containing instructions for generating the architecture, and the generative AI generates the architecture based on the prompt. Step 4: The training data optimization unit proposes methods to optimize the training data using the generative AI. For example, the generative AI adjusts the balance of the data and builds an optimal dataset to improve the model's performance. The input to the generative AI is a prompt containing instructions for optimizing the training data, and the generative AI optimizes the data based on the prompt. Step 5: The hyperparameter tuning part uses the generative AI to automatically tune the model's hyperparameters to find optimal settings. For example, the generative AI adjusts hyperparameters such as the learning rate and batch size to maximize model performance. The generative AI receives inputs from prompts containing instructions for tuning the hyperparameters, and the generative AI adjusts the hyperparameters based on the prompts. Step 6: The automatic evaluation unit uses the generative AI to automatically evaluate the model's performance. For example, the generative AI measures the model's precision and recall using test data and reports the results. The generative AI receives input from a prompt containing instructions for model evaluation, and the generative AI evaluates the model based on the prompt. Step 7: The fine-tuner uses the generative AI to fine-tune the model to improve its performance. For example, the generative AI performs fine-tuning to improve the model's adaptability to a specific task. The input to the generative AI is a prompt containing fine-tuning instructions, and the generative AI fine-tunes the model based on the prompt. Step 8: The model compression unit uses the generative AI to compress the size of the model, enabling efficient execution. For example, the generative AI may reduce unnecessary parameters to make the model lighter. The input to the generative AI is a prompt containing instructions for model compression, and the generative AI compresses the model based on the prompt. Step 9: The continuous learning component uses the generative AI to help the model continuously learn from new data. For example, the generative AI can incorporate new data and suggest ways to update the model. The input to the generative AI is a prompt containing instructions for continuous learning, and the generative AI continuously learns based on the prompt.

[0132] 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.

[0133] 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> Examples of generative AIs include the data generation model 58, such as a neural network model (e.g., a neural network model), and a neural network model (e.g., a neural network model). 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 speech, text data indicating text, and image data indicating an image is also input to the data generation model 58. 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. The specification processing unit 290 performs the above-mentioned specification processing using the data generation model 58. The data generation model 58 may be a fine-tuned model so as to output an inference result from a prompt that does not include an instruction. In this case, the data generation model 58 can output an inference result from a prompt that does not include an instruction. The data processing device 12 and the like include multiple types of data generation models 58, and the data generation model 58 includes AIs other than the generative AI. The AI ​​other than the generative AI may be, for example, linear regression, logistic regression, decision tree, random forest, support vector machine (SVM), k-means clustering, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), or naive Bayes, and can perform various processes, but is not limited to these examples. The AI ​​may also be an AI agent. When the processes of each of the above-mentioned parts are performed by AI, the processes may be performed in part or entirely by AI, but are not limited to these examples. The processes performed by AI, including the generative AI, may be replaced with rule-based processes.

[0134] Furthermore, the processing by the data processing system 10 described above is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart device 14, but may also be executed by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart device 14. Furthermore, the specific processing unit 290 of the data processing device 12 acquires or collects information necessary for processing from the smart device 14 or an external device, and the smart device 14 acquires or collects information necessary for processing from the data processing device 12 or an external device.

[0135] [Second embodiment] FIG. 3 shows an example of the configuration of a data processing system 210 according to the second embodiment.

[0136] 3, the data processing system 210 includes a data processing device 12 and smart glasses 214. An example of the data processing device 12 is a server.

[0137] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, RAM 30, and 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 and / or a LAN.

[0138] 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.

[0139] The microphone 238 receives instructions and the like from the user by receiving voice uttered by the user. The microphone 238 captures the voice uttered by the user, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to instructions from the processor 46.

[0140] 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 user's surroundings (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0141] 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.

[0142] 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.

[0143] 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.

[0144] 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. The identification processing unit 290 can estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0145] In the smart glasses 214, the specific processing is performed by the processor 46. A specific processing program 60 is stored in the storage 50. The processor 46 reads the specific processing program 60 from the storage 50 and executes the read specific processing program 60 on the RAM 48. The specific processing is realized by the processor 46 operating as the control unit 46A in accordance with the specific processing program 60 executed on the RAM 48. Note that the smart glasses 214 may have a data generation model and an emotion identification model similar to the data generation model 58 and the emotion identification model 59.

[0146] Note that a device other than the data processing device 12 may have the data generation model 58. For example, a server device may have the data generation model 58. In this case, the data processing device 12 communicates with the server device having the data generation model 58 to obtain a processing result (such as a prediction result) using the data generation model 58. Furthermore, the data processing device 12 may be a server device, or may be a terminal device (for example, a mobile phone, a robot, a home appliance, etc.) owned by a user.

[0147] 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.

[0148] The data generation model 58 is a so-called generative AI. An example of the data generation model 58 is a generative AI such as ChatGPT. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 receives a prompt containing an instruction, as well as inference data such as voice data representing speech, text data representing text, and image data representing an image. 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. The identification processing unit 290 performs the above-mentioned identification processing using the data generation model 58. The data generation model 58 may be a fine-tuned model so as to output an inference result from a prompt that does not include an instruction. In this case, the data generation model 58 can output an inference result from a prompt that does not include an instruction. The data processing device 12 and the like include multiple types of data generation models 58, and the data generation model 58 includes AIs other than the generative AI. The AI ​​other than the generative AI may be, for example, linear regression, logistic regression, decision tree, random forest, support vector machine (SVM), k-means clustering, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), or naive Bayes, and can perform various processes, but is not limited to these examples. The AI ​​may also be an AI agent. When the processes of each of the above-mentioned parts are performed by AI, the processes may be performed in part or entirely by AI, but are not limited to these examples. The processes performed by AI, including the generative AI, may be replaced with rule-based processes.

[0149] The data processing system 210 according to the second embodiment performs the same processing as the data processing system 10 according to the first embodiment. The processing by the data processing system 210 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the smart glasses 214, but may also be executed by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the smart glasses 214. Furthermore, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the smart glasses 214 or an external device, etc., and the smart glasses 214 acquires or collects information required for processing from the data processing device 12 or an external device, etc.

[0150] [Third embodiment] FIG. 5 shows an example of the configuration of a data processing system 310 according to the third embodiment.

[0151] 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.

[0152] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, RAM 30, and 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 and / or a LAN.

[0153] 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.

[0154] The microphone 238 receives instructions and the like from the user by receiving voice uttered by the user. The microphone 238 captures the voice uttered by the user, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to instructions from the processor 46.

[0155] 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 user's surroundings (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0156] 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.

[0157] 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.

[0158] 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.

[0159] 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. The identification processing unit 290 can estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0160] In the headset type terminal 314, the identification process is performed by the processor 46. A identification program 60 is stored in the storage 50. The processor 46 reads the identification program 60 from the storage 50 and executes the read identification program 60 on the RAM 48. The identification process is realized by the processor 46 operating as a control unit 46A in accordance with the identification program 60 executed on the RAM 48. Note that the headset type terminal 314 may also have a data generation model and an emotion identification model similar to the data generation model 58 and the emotion identification model 59.

[0161] Note that a device other than the data processing device 12 may have the data generation model 58. For example, a server device may have the data generation model 58. In this case, the data processing device 12 communicates with the server device having the data generation model 58 to obtain a processing result (such as a prediction result) using the data generation model 58. Furthermore, the data processing device 12 may be a server device, or may be a terminal device (for example, a mobile phone, a robot, a home appliance, etc.) owned by a user.

[0162] 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.

[0163] The data generation model 58 is a so-called generative AI. An example of the data generation model 58 is a generative AI such as ChatGPT. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 receives a prompt containing an instruction, as well as inference data such as voice data representing speech, text data representing text, and image data representing an image. 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. The identification processing unit 290 performs the above-mentioned identification processing using the data generation model 58. The data generation model 58 may be a fine-tuned model so as to output an inference result from a prompt that does not include an instruction. In this case, the data generation model 58 can output an inference result from a prompt that does not include an instruction. The data processing device 12 and the like include multiple types of data generation models 58, and the data generation model 58 includes AIs other than the generative AI. The AI ​​other than the generative AI may be, for example, linear regression, logistic regression, decision tree, random forest, support vector machine (SVM), k-means clustering, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), or naive Bayes, and can perform various processes, but is not limited to these examples. The AI ​​may also be an AI agent. When the processes of each of the above-mentioned parts are performed by AI, the processes may be performed in part or entirely by AI, but are not limited to these examples. The processes performed by AI, including the generative AI, may be replaced with rule-based processes.

[0164] The data processing system 310 according to the third embodiment performs the same processing as the data processing system 10 according to the first embodiment. The processing by the data processing system 310 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the headset type terminal 314, but may also be executed by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the headset type terminal 314. Furthermore, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the headset type terminal 314 or an external device, etc., and the headset type terminal 314 acquires or collects information required for processing from the data processing device 12 or an external device, etc.

[0165] [Fourth embodiment] FIG. 7 shows an example of the configuration of a data processing system 410 according to the fourth embodiment.

[0166] 7, the 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.

[0167] The data processing device 12 includes a computer 22, a database 24, and a communication I / F 26. The computer 22 includes a processor 28, a RAM 30, and a storage 32. The processor 28, RAM 30, and 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 and / or a LAN.

[0168] 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.

[0169] The microphone 238 receives instructions and the like from the user by receiving voice uttered by the user. The microphone 238 captures the voice uttered by the user, converts the captured voice into audio data, and outputs it to the processor 46. The speaker 240 outputs audio according to instructions from the processor 46.

[0170] 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 image sensor or a CCD image sensor, and captures images of the user's surroundings (for example, an imaging range defined by an angle of view equivalent to the field of vision of a typical healthy person).

[0171] 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.

[0172] The control object 443 includes a display device, LEDs in the eyes, and motors that drive 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.

[0173] 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.

[0174] 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.

[0175] 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. The identification processing unit 290 can estimate the user's emotion using the emotion identification model 59 and perform identification processing using the user's emotion.

[0176] In the robot 414, the processor 46 performs the identification process. A identification program 60 is stored in the storage 50. The processor 46 reads the identification program 60 from the storage 50 and executes the read identification program 60 on the RAM 48. The identification process is realized by the processor 46 operating as a control unit 46A in accordance with the identification program 60 executed on the RAM 48. The robot 414 may have a data generation model and an emotion identification model similar to the data generation model 58 and the emotion identification model 59.

[0177] Note that a device other than the data processing device 12 may have the data generation model 58. For example, a server device may have the data generation model 58. In this case, the data processing device 12 communicates with the server device having the data generation model 58 to obtain a processing result (such as a prediction result) using the data generation model 58. Furthermore, the data processing device 12 may be a server device, or may be a terminal device (for example, a mobile phone, a robot, a home appliance, etc.) owned by a user.

[0178] 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.

[0179] The data generation model 58 is a so-called generative AI. An example of the data generation model 58 is a generative AI such as ChatGPT. The data generation model 58 is obtained by performing deep learning on a neural network. The data generation model 58 receives a prompt containing an instruction, as well as inference data such as voice data representing speech, text data representing text, and image data representing an image. 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. The identification processing unit 290 performs the above-mentioned identification processing using the data generation model 58. The data generation model 58 may be a fine-tuned model so as to output an inference result from a prompt that does not include an instruction. In this case, the data generation model 58 can output an inference result from a prompt that does not include an instruction. The data processing device 12 and the like include multiple types of data generation models 58, and the data generation model 58 includes AIs other than the generative AI. The AI ​​other than the generative AI may be, for example, linear regression, logistic regression, decision tree, random forest, support vector machine (SVM), k-means clustering, convolutional neural network (CNN), recurrent neural network (RNN), generative adversarial network (GAN), or naive Bayes, and can perform various processes, but is not limited to these examples. The AI ​​may also be an AI agent. When the processes of each of the above-mentioned parts are performed by AI, the processes may be performed in part or entirely by AI, but are not limited to these examples. The processes performed by AI, including the generative AI, may be replaced with rule-based processes.

[0180] The data processing system 410 according to the fourth embodiment performs the same processing as the data processing system 10 according to the first embodiment. The processing by the data processing system 410 is executed by the specific processing unit 290 of the data processing device 12 or the control unit 46A of the robot 414, but may also be executed by the specific processing unit 290 of the data processing device 12 and the control unit 46A of the robot 414. Furthermore, the specific processing unit 290 of the data processing device 12 acquires or collects information required for processing from the robot 414 or an external device, etc., and the robot 414 acquires or collects information required for processing from the data processing device 12 or an external device, etc.

[0181] 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.

[0182] 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 encompasses both emotions 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.

[0183] 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.

[0184] 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).

[0185] 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 expressed, and when they approach the ideal, a state of pleasure is expressed. Emotions can also be created for robots, cars, 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 expressed, and when they approach the ideal, a state of pleasure is expressed. An emotion map can be generated, for example, based on Dr. Mitsuyoshi's emotion map (Research on speech emotion recognition and brain physiological signal analysis systems for emotions, Tokushima University, doctoral dissertation: https: / / ci.nii.ac.jp / naid / 500000375379). The left half of the emotion map lists emotions belonging to the area called "reaction," where sensation is dominant. The right half of the emotion map lists emotions belonging to the area called "situation," where situational awareness is dominant.

[0186] 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."

[0187] 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.

[0188] In the above embodiment, an example was given in which a specific process is performed by one computer 22, but the technology disclosed herein is not limited to this, and distributed processing of the specific process may be performed by multiple computers including computer 22.

[0189] 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.

[0190] 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.

[0191] 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.

[0192] The hardware resource for executing a specific process can be any of the following types of processors: A processor, for example, 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. A processor also includes 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.

[0193] The hardware resource that executes the specific process 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 process may be a single processor.

[0194] 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.

[0195] 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.

[0196] In the above example, the first to fourth embodiments have been described separately, but some or all of these embodiments may be combined. The smart device 14, smart glasses 214, headset terminal 314, and robot 414 are merely examples, and they may be combined, or other devices may be used. In the above example, the first and second embodiments have been described separately, but they may be combined.

[0197] 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.

[0198] 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. [Explanation of symbols]

[0199] 10, 210, 310, 410 Data Processing Systems 12 Data Processing Device 14 Smart Devices 214 Smart Glasses 314 Headset-type terminal 414 Robot

Claims

1. A data augmentation part using generative AI, A data cleaning department that uses generative AI, An architecture generation unit that uses generative AI, A training data optimization unit using generative AI, A hyperparameter tuning part using generative AI, An automatic evaluation section using generative AI, A fine-tuning section using generative AI, A model compression part that uses generative AI, A continuous learning unit that uses generative AI. A system characterized by:

2. The data extension unit Using the generative AI, text data with specific emotions is generated, and a dataset based on the emotions is constructed.

2. The system of claim 1.

3. The data cleaning unit The generative AI is used to check the consistency of the data and automatically correct any inconsistencies in the data.

2. The system of claim 1.

4. The architecture generation unit Using the generative AI, the optimal architecture for different tasks is automatically generated, and a model optimized for each task is provided.

2. The system of claim 1.

5. The training data optimization unit Using the generative AI to optimize training data based on emotions and train a model that adapts to the emotions.

2. The system of claim 1.

Images