Method and system for artificial intelligence inference

Abstract

The present invention relates to a method and a system for artificial intelligence inference, and provides a method and a system for artificial intelligence inference, which are for solving various problems through inference by using artificial intelligence.

Description

Artificial Intelligence Reasoning Methods and Systems

[0001] The present invention relates to an artificial intelligence reasoning method and system, and provides an artificial intelligence reasoning method and system that solves various problems through reasoning using artificial intelligence.

[0002] The dictionary definition of artificial intelligence is a technology that realizes human learning, reasoning, perception, and natural language understanding abilities through computer programs. This artificial intelligence has achieved rapid development through deep learning.

[0003] Driven by such advancements in artificial intelligence, research aimed at actively utilizing AI in various fields is actively underway. More specifically, as AI technology develops, research on applying AI to diverse target tasks is being actively pursued.

[0004] In particular, the problem of achieving specific goals while considering various constraints is recognized as a significant challenge in various industries and service sectors (e.g., logistics, manufacturing, network design, resource allocation, scheduling, pathfinding, etc.).

[0005] In this regard, combinatorial optimization (CO) problems, which involve searching for an optimal solution that minimizes cost or maximizes performance under various constraints, are utilized in various industries and fields (e.g., path optimization, resource allocation, scheduling, graph-based selection problems, etc.).

[0006] These combinatorial optimization problems are characterized by a rapid increase in search space as the problem scale or the number of constraints increases, making it difficult for humans to derive the optimal solution through direct search.

[0007] In addition, even problems of the same type can be defined as different forms of problems depending on the configuration or combination of applied constraints, and as a result, problem types are subdivided in various ways.

[0008] Current AI-based approaches often construct models specialized for individual problem types or build models based on training data limited to specific combinations, which has limitations in that learned information is not sufficiently shared between different problem combinations.

[0009] Accordingly, if the combination of constraints changes, retraining of the model may be required, or the consistency of prediction results may decrease for new, unlearned problem combinations.

[0010] Due to these characteristics, technical considerations regarding the scope of application and generalization performance of artificial intelligence models are required in environments where various problems exist.

[0011] The present invention aims to provide an artificial intelligence reasoning method and system that solves various problems through reasoning using artificial intelligence.

[0012] In particular, the present invention is intended to provide an artificial intelligence reasoning method and system capable of flexibly responding to various problems.

[0013] Furthermore, the present invention aims to provide an artificial intelligence model that can be universally applied to various problems and utilized effectively to solve various problems.

[0014] More specifically, the present invention aims to provide an artificial intelligence model that is universally applicable to various problems and capable of reasoning even on new problems that have not been learned.

[0015] Furthermore, the present invention aims to provide an artificial intelligence reasoning method and system capable of efficiently solving problems occurring in various environments through reasoning using artificial intelligence.

[0016] To solve the problem described above, the artificial intelligence inference method according to the present invention, which is performed by a computer, presents a method for generating prediction results for various problems using a learned prediction model. Furthermore, the present invention provides a processing pipeline that interprets a problem including multiple attributes into an attribute-unit information processing structure and generates prediction results based thereon.

[0017] More specifically, the inference method of artificial intelligence according to the present invention may include the steps of receiving a prediction request, inputting the prediction request to a prediction model trained such that a representation vector corresponding to a plurality of attributes is reflected in an embedding, and generating a prediction result based on at least one specific attribute forming the prediction request by the prediction model.

[0018] In an embodiment, the representation vector corresponding to the plurality of attributes is generated during the learning process of the prediction model based on the difference between the first embedding data and the second embedding data, and the first embedding data and the second embedding data can each be generated by data distinguished according to whether or not at least one attribute is included.

[0019] In an embodiment, the second embedding data may be embedding data generated by inputting the data from which at least one attribute has been removed into the prediction model.

[0020] In an embodiment, the prediction model can be trained so that the representation vectors have similar representations in a vector space.

[0021] In the embodiment, the representation vector may be a vector corresponding to the same attribute included in different attribute combinations.

[0022] In an embodiment, the representation vector can be combined as part of the embedding to form an embedding component corresponding to the specific attribute.

[0023] In an embodiment, the embedding may include a first embedding component representing a unique meaning corresponding to each attribute and a second embedding component related to the interaction of a plurality of attributes.

[0024] In an embodiment, the embedding may be formed by the combination of the first embedding component and the second embedding component.

[0025] In an embodiment, the prediction result may be generated based on the first embedding component corresponding to the specific attribute.

[0026] In an embodiment, even if the combination of attributes including at least one specific attribute is not used in the learning process of the prediction model, the prediction model can generate a prediction result based on a representation vector corresponding to the at least one specific attribute.

[0027] In an embodiment, the expression vector may be formed to represent a meaning corresponding to the specific attribute independently of other attribute combinations that include the specific attribute.

[0028] In an embodiment, the prediction model is trained to output a solution to a combinatorial optimization problem as the prediction result, the combinatorial optimization problem includes at least one of a resource placement, path determination, or schedule establishment problem combined with a plurality of constraints, and the plurality of attributes may include attributes defining the constraints.

[0029] A learning method for artificial intelligence according to the present invention may include the steps of: receiving data containing attribute information corresponding to each of a plurality of attributes; inputting the data into a prediction model to generate first embedding data; inputting data from which at least one attribute has been removed into the prediction model to generate second embedding data; and training the prediction model such that a representation vector corresponding to the at least one attribute is formed based on the difference between the first embedding data and the second embedding data.

[0030] In an embodiment, the first embedding data and the second embedding data may be generated based on whether the at least one attribute is included for the same data.

[0031] In an embodiment, the second embedding data may be generated by inputting the data from which at least one attribute has been removed into the prediction model.

[0032] In an embodiment, the prediction model can be trained so that the representation vectors corresponding to the same attribute included in different attribute combinations have similar representations in a vector space.

[0033] In an embodiment, the prediction model can be trained such that a first embedding component representing a unique meaning corresponding to each attribute and a second embedding component corresponding to the interaction of a plurality of attributes are formed separately.

[0034] In an embodiment, the first embedding data can be learned to be formed by the combination of the first embedding component and the second embedding component.

[0035] An artificial intelligence inference system according to the present invention comprises one or more processors configured to perform operations by executing one or more instructions and a memory connected to said processors, wherein a prediction model learned such that a representation vector corresponding to a plurality of attributes is reflected in an embedding is stored in said memory, and said processors may be configured to receive a prediction request, input said prediction request into said prediction model, and generate a prediction result based on at least one specific attribute among a plurality of attributes constituting said prediction request.

[0036] A program according to the present invention is a program that is executed by one or more processes in an electronic device and can be stored on a computer-readable recording medium, and may include instructions for performing the steps of receiving a prediction request, inputting the prediction request into a prediction model learned such that a representation vector corresponding to a plurality of attributes is reflected in an embedding, and generating a prediction result based on at least one specific attribute forming the prediction request by the prediction model.

[0037] As described above, according to the artificial intelligence inference method and system of the present invention, the prediction model according to the present invention is trained to form attribute-specific representation vectors based on the difference between first embedding data and second embedding data, thereby enabling the precise extraction of the unique meaning of each attribute by separating it from other attribute combinations. Accordingly, the meaning of the same attribute is consistently maintained even between different problem combinations, and stable prediction results can be generated even for new attribute combinations that have not been learned. Accordingly, the prediction accuracy and generalization performance of the prediction model regarding combination optimization problems can be improved.

[0038] In addition, according to the artificial intelligence inference method and system of the present invention, by separating and configuring a first embedding component representing the intrinsic meaning of each attribute and a second embedding component corresponding to the interaction of multiple attributes, the prediction model is enabled to learn by structurally decomposing the intrinsic meaning and contextual interaction of attributes. Through this, the prediction model is trained to maintain similar representations in a vector space even when the same attribute is included in different attribute combinations, and can consistently preserve analogy relationships between attributes in the embedding space. As a result, knowledge sharing between problem types can be efficiently achieved, and high performance can be achieved with a single model for various modified problems.

[0039] Furthermore, according to the artificial intelligence inference method and system of the present invention, a prediction model is trained such that embedding transformations resulting from the addition or removal of specific attributes have similar directionality even in different problem contexts, thereby enabling the formation of analogous relationships between attributes in the embedding space. Through this, the prediction model can secure generalization performance capable of generating accurate prediction results even for problems involving new combinations that have not been learned. Therefore, the present invention can flexibly respond to combinations of constraints that continuously change in actual industrial environments.

[0040] Furthermore, the artificial intelligence inference method and system according to the present invention can be applied to combinatorial optimization problems involving the combination of multiple constraints. By decomposing each constraint into attribute units and vectorizing their representations, it is possible to derive stable and precise solutions even in complex constraint environments. In other words, the present invention can simultaneously improve the accuracy of problem solving and generalization performance in real-world logistics, transportation, and scheduling environments where various constraints change dynamically.

[0041] Furthermore, according to the artificial intelligence inference method and system of the present invention, by structuring attribute representations into a reusable form, rapid adaptation with a small amount of data is possible even when new attributes or combinations of new attributes are added. Through this, the present invention can respond to new problem types solely through efficient parameter adjustment without the need to retrain the entire model from scratch, thereby reducing training costs and time resources.

[0042] FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented.

[0043] FIG. 2 illustrates an example of a block diagram of a computing device that may be included in a user computing device, a server computing system, and a training computing system, as an embodiment of a computing system in which the present invention can be implemented.

[0044] Figure 3 illustrates an example of a block diagram from another perspective of a computing device, which is one of the components of a computing system.

[0045] FIG. 4 is a conceptual diagram illustrating an artificial intelligence inference system according to the present invention.

[0046] FIGS. 5, FIGS. 6, and FIGS. 7 are conceptual diagrams for explaining a prediction model according to the present invention.

[0047] FIG. 8 illustrates examples of attributes according to the present invention.

[0048] FIG. 9a is a flowchart illustrating an artificial intelligence reasoning method according to the present invention.

[0049] FIG. 9b is a flowchart illustrating a learning method of artificial intelligence according to the present invention.

[0050] FIGS. 10, FIGS. 11, FIGS. 12, FIGS. 13, FIGS. 14, FIGS. 15, FIGS. 16, FIGS. 17, FIGS. 18, FIGS. 19, FIGS. 20, FIGS. 21, FIGS. 22, FIGS. 23, FIGS. 24, FIGS. 25, FIGS. 26, FIGS. 27, FIGS. 28, FIGS. 29, FIGS. 30, and FIGS. 31 are drawings illustrating conceptual diagrams and mathematical expressions for explaining a learning and inference method of artificial intelligence according to the present invention.

[0051] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components are assigned the same reference number regardless of the drawing symbols, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not have distinct meanings or roles in themselves. Furthermore, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the embodiments disclosed in this specification, such detailed description will be omitted. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings, and it should be understood that they include all modifications, equivalents, and substitutions that fall within the spirit and technical scope of the present invention.

[0052] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. These terms are used solely for the purpose of distinguishing one component from another.

[0053] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0054] Singular expressions include plural expressions unless the context clearly indicates otherwise.

[0055] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0056] Meanwhile, FIG. 1 illustrates an example of a block diagram of a computing system in which the present invention can be implemented. In this regard, the inference system of artificial intelligence according to the present invention can be implemented through a computing device described below and can perform data processing related to the inference method of artificial intelligence described in this specification.

[0057] Referring to FIG. 1, a computing system (10000) that performs a method of solving various problems through reasoning using artificial intelligence according to one embodiment of the present invention may include at least one computing device. At this time, the at least one computing device may be a single processor or a multi-processor computing device.

[0058] The components of at least one computing device of the present invention may include various hardware components such as one or more processors, memory, other hardware, and a system bus (not shown) that connects various system components so that they can transmit and receive data to and from each other (e.g., telecommutatively connected, physically connected, electrically connected), and the components of at least one computing device are not limited thereto and may be very diverse.

[0059] Meanwhile, at least one computing device included in a computing system (10000) that performs a method of solving various problems through reasoning using artificial intelligence may be connected to communicate via a network (1070). For example, at least one computing device included in the computing system (10000) may be clustered or may be part of a local area network (LAN). Additionally, at least one computing device may be part of a wide area network (WAN) or connected to at least one of a client-server network and a peer-to-peer network within the cloud.

[0060] Meanwhile, when at least one computing device is used in at least one of a network environment and a cloud computing environment, the at least one computing device may be connected to at least one of a public and a private network through a network interface or an adapter. In one embodiment, other communication connection devices, such as a modem, may be used to establish communication through the network. The modem may be at least one of an internal modem and an external modem, and may be connected to a system bus through a network interface or a specific mechanism, etc. A wireless network component consisting of an interface and an antenna may be coupled to the network through a device such as an access point, a peer computer, etc. In the present invention, the method by which at least one computing device is connected to communicate through the network (1070) is not limited and may be connected to communicate in a manner different from the described examples.

[0061] Furthermore, other computer-type devices or systems not shown in FIG. 1 may also interact technically with at least one computing device or other system through one or more connections to the network (1070) via a network interface. Here, the network interface may include network interface equipment such as a physical network interface controller (NIC) or a virtual network interface (VIF).

[0062] The network (1070) of the present invention may include various forms such as the Internet, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wireless USB (Wireless Universal Serial Bus), etc., and in the present invention, data transmission may be performed based on standard communication protocols such as TCP / P, HTTP, SSL, etc.

[0063] A computing system (10000) that performs a method of solving various problems through reasoning using artificial intelligence according to the present invention may include at least one of a user computing device (1010), a training computing system (1050), and a server computing system (1030).

[0064] A user computing device (1010) according to the present invention may be understood as a computing device comprising at least one processor (1011) and at least one memory (1012) that perform a method of solving various problems through reasoning using artificial intelligence. For example, the user computing device (1010) may include at least one computing device among a smartphone, a smart TV, a laptop computer, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, a slate PC, a tablet PC, an ultrabook, a wearable device (e.g., a smartwatch, a smart glass, and a head-mounted display).

[0065] At least one or more processors (1011) constituting the user computing device (1010) may include one or more general-purpose processors and one or more special-purpose processors. For example, at least one or more processors (1011) constituting the user computing device (1010) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), an application integrated circuit, an application semiconductor (ASIC), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and other electrical units for performing functions.

[0066] Furthermore, at least one or more processors (1011) may be configured to execute computer-readable instructions contained in memory (1012) or other instructions described herein.

[0067] The memory (1012) constituting the user computing device (1010) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media, or other types of physically durable storage media.

[0068] For example, the memory (1012) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, and combinations thereof, and may include web storage of a server that performs the storage function of memory on the internet. This memory (1012) may store data and instructions necessary for the at least one or at least one processor (1011) to perform the operation of an application for solving various problems through reasoning using artificial intelligence.

[0069] A user computing device (1010) may include one or more user input components (1021) that detect user input. For example, the user input component (1021) may also be referred to as a user interface module. The user input component (1021) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of user input component (1021). In this case, the user input component (1021) in the present invention does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user. Meanwhile, the user of the present invention may refer to an automated agent, script, playback software, etc., that operates on behalf of one or more people.

[0070] A user can interact with a computing system (10000) including at least one computing device through input text, touch, voice, movement, computer vision, gestures, and other forms of input / force using a user input component (1021). For example, the user input component (1021) may include one or more of a command line interface (CLI), a graphical user interface (GUI), a natural user interface (NUI), a voice command interface, and other user interface (UI) representations.

[0071] Between the user input component (1021) and the user computing device (1010), one or more application programming interface (API) calls may be made based on user input received from a user interface or network.

[0072] Here, the expression "based on" may be interpreted to include cases where it is based on the use of a specific configuration, modified from, derived from, influenced by, dependent on, or otherwise derived from a specific configuration. In some embodiments, an API call may be configured for a specific API, which may be interpreted or converted into an API call configured for another API. Here, an API may refer to a defined interface or connection between computers or between computer programs.

[0073] In one embodiment, the user computing device (1010) may store at least one machine learning model (1020). For example, the user computing device (1010) may be various machine learning models, such as a plurality of neural networks (e.g., deep neural networks) that perform problem solving through artificial intelligence-based reasoning using at least one problem and at least one specific attribute constituting the problem, or other types of machine learning models including non-linear models or linear models, and may be composed of a combination thereof.

[0074] According to an embodiment of the present invention, a user computing device (1010) can perform a problem-solving method through artificial intelligence-based inference by using a local or / or external machine learning model (1020). Alternatively, the user computing device (1010) may perform a problem-solving method through artificial intelligence-based inference by using a machine learning model (1040) provided by a server.

[0075] In addition, according to another embodiment of the present invention, a server computing system (1030) communicating with a user computing device (1010) may provide a prediction result for at least one problem to the user computing device (1010) on an application or / on the web in accordance with a request from a user received through the user computing device (1010).

[0076] In addition, according to another embodiment of the present invention, by linking at least a part of a user computing device (1010) and a server computing system (1030) with each other and performing a method to solve various problems through reasoning using artificial intelligence, a prediction result for at least one problem can be provided to the user.

[0077] Additionally, according to various embodiments of the present invention, a user computing device (1010) or a server computing system (1030) can learn machine learning models (1020, 1040) performed in a problem-solving method using artificial intelligence-based inference through interaction with a training computing system (1050) that is communicatedly connected via a network (1070). In this case, the training computing system (1050) may be a computing system separate from the server computing system (1030). Alternatively, in some embodiments, the training computing system (1050) may be part of the server computing system (1030) or part of the user computing device (1010).

[0078] Meanwhile, the server computing system (1030) may include at least one processor (1031) and memory (1032). Here, the processor (1031) may be composed of at least one or a plurality of electrically connected processors among a central processing unit (CPU), a graphics processing unit (GPU), a tensor processing unit (TPU), a neural network processing unit (NPU), an application integrated circuit, an application semiconductor (ASIC), an arithmetic logic unit (ALU), a floating-point arithmetic unit (FPU), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, and other electrical units for performing functions. For example, at least one processor (1031) may include a circuit and a transistor configured to execute instructions from memory (1032).

[0079] The memory (1032) constituting the server computing system (1030) according to the present invention may include volatile memory, non-volatile memory, fixed media, removable media, magnetic media, optical media, semiconductor media, and other types of physically durable storage media. For example, the memory (1032) may include one or more non-transient / transient computer-readable storage media such as RAM, ROM, HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), EEPROM, EPROM, flash memory device, magnetic disk, etc., and combinations thereof, and may include web storage of a server that performs the storage function of memory over the internet. Additionally, the server computing system (1030) may further include a data storage (data store). For example, the data storage may be composed of at least one of a relational database, a NoSQL database, a data warehouse, and a local file system.

[0080] In the memory (1032) constituting the server computing system (1030) according to the present invention, data and instructions necessary for the at least one processor (1031) to perform the operation of an application for solving various problems through reasoning using artificial intelligence may be stored.

[0081] In one embodiment, the server computing system (1030) may be composed of a single device or a plurality of computing devices, and may be configured to operate according to a sequential or parallel computing architecture. Additionally, a distributed processing system may be configured with a plurality of networked devices.

[0082] Meanwhile, the training computing system (1050) may include at least one processor (1051) and memory (1052). The model trainer (1060) is a logical component that executes the training of at least one machine learning model (1020, 1040) and may be implemented in the form of hardware, firmware, or software. For example, the model trainer (1060) may be executed by the processor (1051) after loading training data (1061) stored in a storage device into memory (1052). For example, the model trainer (1060) may be configured to execute one or more operations (e.g., model training, model reconstruction, model validation, model testing) on ​​at least one machine learning model.

[0083] The machine learning model of the present invention may include at least one of a statistical model, an algorithm, a neural network (NN), a convolutional neural network (CNN), a generative neural network (GNN), a Word2Vec model, a Bag of Words model, a TF-IDF (document frequency-inverse document frequency) model, a GPT (Generative Pre-trained Transformer) model (or other autoregressive models), a PPO (Proximal Policy Optimization) model, a nearest neighbor model (e.g., a k-nearest neighbor model), a linear regression model, a K-means clustering model, a Q-learning model, a TD (Temporal Difference) model, a Deep Adversarial Network model, and all other types of models further described herein.

[0084] Specifically, the model trainer (1060) may execute operations to train a machine learning model, and said operations may include at least one of adding, removing, and modifying model parameters. At this time, the training of the machine learning model may be at least one of supervised learning, semi-supervised learning, and unsupervised learning. In one embodiment, the training of the machine learning model may include the step of repeatedly inputting training data (1061) based on epochs and repeatedly performing the machine learning model training process configured in this way. Here, an epoch may refer to a unit in which the entire set of training data (1061) undergoes forward and backpropagation processing once. In some implementations, different levels of training methods (e.g., supervised learning, semi-supervised learning, unsupervised learning) may be used for different epochs.

[0085] The training data (1061) of the present invention may include input data and data previously output from at least one machine learning model (e.g., recursive learning feedback).

[0086] At least one parameter of a machine learning model may include at least one of a seed value, a model node, a model layer, an algorithm, a function, connections between different machine learning models, connections between parameters, machine learning model constraints, and other digital components that influence the output of the machine learning model. In this case, model connections between different machine learning models may include or represent model parameters or relationships between models, and may be dependent or interdependent, hierarchical, and / or static or dynamic. The combinations and configurations of model parameters described herein may be too complex to be maintained or utilized by human cognitive abilities.

[0087] In the present invention, the machine learning parameters described according to the embodiments are not limited, and a single machine learning model may further include a plurality of model parameters.

[0088] Meanwhile, FIG. 2 illustrates an example of a block diagram of a computing device (1100) that may be included in a user computing device (1010), a server computing system (1030), and a training computing system (1050), as an embodiment of a computing system (10000) in which the present invention can be implemented.

[0089] As illustrated in FIG. 2, the computing device (1100) may include at least one application (e.g., Application 1 to Application N), and each of the at least one application may include a machine learning library and a model execution environment for performing a problem-solving method through machine learning-based inference. The at least one application included in the computing device (1100) may communicate with the sensor, context manager, device state manager, or additional component(s) within the computing device (1100) via an Application Programming Interface (API). In one embodiment, the at least one application may interface with device components, such as receiving sensor data or state data or transmitting prediction results to an output device via a public or private API.

[0090] Meanwhile, FIG. 3 illustrates an example of a block diagram in another aspect of a computing device (1200), which is one of the components of a computing system (10000) that performs a method of solving various problems through reasoning using artificial intelligence according to an embodiment of the present invention.

[0091] A computing device (1200) according to the present invention may include at least one application (e.g., Application 1 to Application N), and at least one application may communicate with a central intelligence layer (1210). Each application may interact with a shared model within the central intelligence layer (1210) through an API (e.g., a common API).

[0092] The central intelligence layer (1210) includes one or more machine learning models and may share them among multiple applications or provide them independently to each. In one embodiment, the central intelligence layer (1210) may be integrated as part of an operating system or implemented as a separate logical layer.

[0093] Additionally, the central intelligence layer (1210) can communicate with the central device data layer (1220). The central device data layer (1220) can store at least one problem and attributes included in said problem within the computing device (1200) and provide them as input data necessary to solve various problems through reasoning using artificial intelligence. Each device component (e.g., sensor, state manager, etc.) can communicate with the central device data layer (1220) through a private API, etc.

[0094] The technology described in this specification may be composed of a single or multiple computing devices, and a machine learning model that performs a method of solving various problems through reasoning using artificial intelligence may be executed sequentially or in parallel on a single component or multiple distributed components. Data storage, machine learning models, and applications may be distributed and operated locally or over a network, and these configurations can be flexibly applied to various system architectures.

[0095] Meanwhile, the present invention relates to an artificial intelligence reasoning method and system that solves various problems through reasoning using artificial intelligence. In particular, the artificial intelligence reasoning system according to the present invention may be a system capable of flexibly responding to various problems.

[0096] In this specification, the “problem” may include various types of problems. More specifically, the “problem” described in this specification may include problems that achieve a specific goal by considering various constraints. As an example, the problem may include a combinatorial optimization (CO) problem. A combinatorial optimization problem is a problem of finding an optimal solution that minimizes cost or maximizes performance under various constraints, and is utilized in various industries and fields (e.g., path optimization, resource allocation, scheduling, graph-based selection problems, etc.).

[0097] In this regard, the Vehicle Routing Problem (VRP) is a combinatorial optimization problem in logistics and transportation that seeks an optimal solution (i.e., a driving plan) that minimizes total costs (e.g., total travel distance or total transportation costs) while satisfying various constraints (e.g., vehicle load capacity, driving distance, driving time, etc.) when delivering materials from one or more depots to multiple customers (or demand centers).

[0098] For example, a variation of the vehicle routing optimization problem (or vehicle routing problem), including the Capacitated Vehicle Routing Problem (CVRP), may be defined by constraints corresponding to the activated attributes among the set of attributes described below. An instance of the Capacitated Vehicle Routing Optimization Problem (or Capacitated Vehicle Routing Problem) (see FIG. 18 (a)) is defined on a graph (G) having a set of nodes (see FIG. 18 (b)), and edge weights may be given by Euclidean distances (see FIG. 18 (c)). Here, node 0 represents a depot (start and return point), and the remaining nodes may represent customers. Each node has coordinates (see FIG. 18 (d)) and an attribute feature (see FIG. 18 (e)) that defines constraints specific to the vehicle routing optimization problem. In this specification, “attribute feature” may also be expressed as “attribute characteristic.”

[0099] The goal of the vehicle path optimization problem is to find an optimal solution (see (f) in Fig. 18), where intermediate depot visits can divide the optimal solution into K paths. Nodes corresponding to every customer must be visited exactly once. In this case, the objective function may be to minimize the total travel distance (see (h) in Fig. 18) while satisfying constraints defined by all attribute features (or all attribute characteristics, see (g) in Fig. 18).

[0100] Additionally, variations of the vehicle path optimization problem may be extended by combining additional active attributes (or activated attributes) based on a capacity-constrained vehicle path optimization problem that includes at least one attribute (e.g., Linehaul, Capacity(Q)), etc. Each variation must satisfy the constraints of both the capacity and the active attributes. In one embodiment, the active attributes may include at least one of Backhaul(B), Mixed Backhaul(MB), Open(O), Time Window(TW), and Linehaul(L).

[0101] For an instance (see FIG. 19 (a)), the active attribute can be represented by a binary attribute indicator vector (see FIG. 19 (b)). The vector for each of the active attributes (or the corresponding vector) can be represented as in FIG. 19 (c). In this case, if attribute (A) is active, the attribute may be included in the problem. On the other hand, if the attribute is not active, the attribute may not be included in the problem. For example, a problem involving a time window (e.g., Open Vehicle Routing Problem, OVRPTW) includes an open and a time window, so the vector for it can be represented as “(0, 0, 1, 1, 0)”.

[0102] Furthermore, the present specification may represent the vehicle path optimization problem as a sequential decision process within a Markov Decision Process (MDP) framework. In this case, the solution may be constructed autoregressively. In this Markov decision process formulation, the state at time t (see Fig. 19 (d)) may consist of an instance (x) and a partial solution (see Fig. 19 (e)). The agent may select the next node (see Fig. 19 (f)) within the range satisfying the constraints. In this case, the reward may be defined as in Fig. 19 (g).

[0103] Here, “configured autoregressively” can also be understood as using an autoregressive policy composed of an encoder and a decoder. Such a policy defines conditional probabilities (see FIG. 20 (a)), and the probability of a solution being generated can be represented as in FIG. 20 (b). In this case, the goal may be to maximize the expected reward (see FIG. 20 (c)). To this end, at least one algorithm (e.g., the REINFORCE algorithm) or at least one technique (e.g., POMO (Policy Optimization with Multiple Optima) technique, variation-specific reward normalization technique, etc.) may be used or applied in this specification.

[0104] Meanwhile, the artificial intelligence reasoning system according to the present invention may include an artificial intelligence model that can be universally applied to various problems and utilized effectively to solve various problems. The present invention aims to efficiently solve various problems by flexibly responding to various problems using such an artificial intelligence model.

[0105] Hereinafter, the inference system of artificial intelligence according to the present invention will be examined in more detail together with the attached drawings. FIG. 4 is a conceptual diagram for explaining the inference system of artificial intelligence according to the present invention, and FIGS. 5, 6, and 7 are conceptual diagrams for explaining the artificial intelligence model according to the present invention. Furthermore, FIG. 8 illustrates an example of attributes according to the present invention.

[0106] Meanwhile, as illustrated in FIG. 4, the artificial intelligence inference system (1000) according to the present invention may include at least one of an input unit (100), an output unit (200), a communication unit (300), a storage unit (400), a data processing unit (500), an artificial intelligence model (600), and a control unit (700). However, the components of the artificial intelligence inference system (1000) according to the present invention are not necessarily limited thereto and may further include various hardware components that perform the same or similar roles as the functions described in the present specification.

[0107] Although not illustrated, the artificial intelligence inference system (1000) according to the present invention may include one or more processors, and such processors may include one or more general-purpose processors and one or more special-purpose processors (e.g., digital signal processor, tensor processing unit (TPU), graphics processing unit (GPU), neural network processing unit (NPU), application integrated circuit, application semiconductor (ASIC), field programmable gate array (FPGA), quantum processing unit (or quantum processor, QPU), etc.). One or more processors may be configured to execute instructions, computer-readable instructions, and other instructions described herein that are stored (or included) in the storage unit (400). The artificial intelligence inference method and system according to the present invention may perform data processing described below in cooperation with memory and at least one processor. The processor may perform a series of operations and data processing using data and information stored in memory. In this case, memory may be a component of the storage unit (400). The above memory may be connected to a processor, and the memory may store a prediction model trained such that representation vectors corresponding to multiple attributes are reflected in the embedding.

[0108] Meanwhile, the input unit (100) can be configured in various ways as a means of data input. For example, the input unit (100) can be configured to receive user input. The input unit (100) can be configured to receive user input from a user terminal. Here, “receiving input” may mean receiving an input signal (or selection signal) corresponding to the user’s input based on input being made by the user through the configuration of the input unit provided in the user terminal.

[0109] Here, the user terminal may include at least one of a mobile phone, a smartphone, a notebook computer, a laptop computer, a slate PC, a tablet PC, an ultrabook, a desktop computer, a digital broadcasting terminal, a PDA (personal digital assistants), a PMP (portable multimedia player), a navigation device, and a wearable device (e.g., a smartwatch, a smart glass, a head-mounted display).

[0110] In addition, the input unit (100) in the present invention does not necessarily mean a hardware means, but can be understood as a channel for receiving input from a user.

[0111] The input unit (100) may also be referred to as a user interface module. The input unit (100) may include a touch screen, a computer mouse, a keyboard, a keypad, a touchpad, a trackball, a joystick, a voice recognition module, or other similar devices. However, the present invention does not limit the type of input unit (100).

[0112] Here, user input may include documents, text, images (or videos), voice, etc. In this case, the artificial intelligence inference system (1000) may further include a module that converts voice into text.

[0113] Next, the output unit (200) can output information through an output unit configuration (e.g., a display unit, a touch screen, a speaker, etc.) provided in a user terminal linked to the artificial intelligence inference system (1000) according to the present invention. For example, the output unit (200) can output at least one page (or service page) linked to the artificial intelligence inference system (1000) according to the present invention to the display unit of the user terminal. Additionally, the output unit (200) does not necessarily mean a hardware means, but can be understood as a channel for outputting results to a user.

[0114] Next, the communication unit (300) may be connected via a wireless or wired network to a user terminal, a server (e.g., a central server, an external server, etc.), a device, and at least one network, etc., to receive or transmit overall data and information necessary for the operation of the artificial intelligence inference system (1000) according to the present invention.

[0115] The communication unit (300) can support various communication methods according to the communication standards of the communicating device. The processor can transmit and receive data with an external server and a user terminal using the communication unit.

[0116] For example, the communication unit (300) may be configured to communicate with a communication target using at least one of the following technologies: WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Wi-Fi (Wireless-Fidelity) Direct, DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband), WiMAX (World Interoperability for Microwave Access), HSDPA (High Speed ​​Downlink Packet Access), HSUPA (High Speed ​​Uplink Packet Access), LTE (Long Term Evolution), LTE-A (Long Term Evolution-Advanced), 5G (5th Generation Mobile Telecommunication), Bluetooth (Bluetooth™), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), UWB (Ultra-Wideband), ZigBee, NFC (Near Field Communication), Wi-Fi Direct, and Wireless USB (Wireless Universal Serial Bus).

[0117] Next, the storage unit (400, or memory) serves to store various data related to the present invention and may include one or more non-transient computer-readable storage media that can be read and accessed by at least one of one or more processors.

[0118] One or more computer-readable storage media may include volatile and non-volatile storage components, such as optical, magnetic, organic, or other memory or disk storage devices. In some examples, the storage unit (400) may be implemented using a single physical device (e.g., one optical, magnetic, organic, or other memory or disk storage device), whereas in other examples, the storage unit (400) may be implemented using two or more physical devices.

[0119] The storage unit (400) may include computer-readable instructions and additional data. The storage unit (400) may include a storage necessary to perform at least some of the methods, scenarios, and techniques described herein and at least some of the functions of the device and network.

[0120] Furthermore, at least a portion of the storage unit (400) may be a cloud storage or a cloud server. At least a portion of the data corresponding to user input received from the input unit (100) and the training data (or training data set) may be stored in the storage unit (400).

[0121] That is, the storage unit (400) is sufficient as a space where information necessary for the operation of the artificial intelligence inference system (1000) according to the present invention is stored, and it can be understood that there are no restrictions on the physical space.

[0122] Furthermore, the storage unit (400) can store a computer program including computer program instructions. Furthermore, the storage unit (400) can store a computer program including computer program instructions that control the operation of the system (1000) or control the operation of the control unit (700) when loaded into the processor of the system (1000).

[0123] Next, the data processing unit (500) may be configured to perform data preprocessing on at least one piece of data used for learning or inference of the artificial intelligence model (600). In this case, the preprocessed data may be processed as input to the artificial intelligence model (600).

[0124] Additionally, the data processing unit (500) can collect (or receive) data (or information) necessary for the operation of the artificial intelligence inference system (1000) from at least one of various sources (e.g., a website, a database (DB), an application programming interface (API) linked to the artificial intelligence inference system (1000), a server linked to the artificial intelligence inference system (1000), a central server, an external server, cloud storage, a user terminal, a large dataset, etc.).

[0125] Furthermore, the data processing unit (500) can process at least one piece of data required for the learning as an input to the artificial intelligence model (600) so that the artificial intelligence model (600) is learned in the attribute unit constituting at least one problem.

[0126] For example, the data processing unit (500) can process a problem (1, or data) containing attribute information (1a, 1b, 1c) corresponding to each of a plurality of attributes as input to the artificial intelligence model (600), thereby enabling the artificial intelligence model (600) to learn the various problems (1, 2, 3) in terms of attribute units. More specific details regarding this will be explained later.

[0127] Next, the artificial intelligence model (600) may be configured to perform the role of solving various problems (1, 2, 3) through inference. In the present invention, the artificial intelligence model (600) may be a model that is universally applicable to various problems (1, 2, 3) and capable of inferencing new problems that have not been learned. In this specification, the artificial intelligence model (600) may also be referred to as a “prediction model.” Additionally, in this specification, the learned artificial intelligence model (600) may also be referred to as a “learned prediction model.”

[0128] Specifically, the artificial intelligence model (600) may be configured to generate a prediction result through inference on at least one piece of data (or input data) corresponding to (or included in) a prediction request. In this case, the data may be data containing attributes (or attribute information) corresponding to each of a plurality of attributes. Alternatively, the data may include a problem containing (or composed of) at least one attribute.

[0129] In one embodiment, in a variation of the vehicle path optimization problem, attributes may have at least one of a first attribute (e.g., Intrinsic Semantics for Individual Attributes) and a second attribute (e.g., Contextual Cross-Attribute Interactions).

[0130] For example, a single attribute has an inherent and intrinsic meaning and can maintain the same constraint definition across all combinations of attributes. That is, a single attribute can have the same intrinsic meaning regardless of the combination. In this specification, such a characteristic of an attribute may be referred to as the “first characteristic of the attribute.” For example, the influence of a specific attribute (e.g., Linehaul(L)) can enforce the same maximum path length limit in both VRPL and OVRPL, which have different problem structures.

[0131] For example, an attribute has a first characteristic, but a combination of attributes can generate contextual interactions that go beyond the effects of the individual attributes. That is, attributes maintain a first characteristic that has inherent meaning, but additional contextual interactions may occur when combined with one another. In this specification, such characteristics of an attribute may be referred to as “second characteristics of the attribute.” For example, the influence of a specific attribute (e.g., Linehaul(L)) can be significantly weakened when present with an open route attribute (e.g., Open(O)). This is because length constraints are significantly relaxed as the return to the depot in OVRPL is removed, thereby reducing the influence of the said specific attribute compared to its role in VRPL.

[0132] In this regard, the encoder in the present invention may decompose the final representation (see (d) of FIG. 20) into at least one component to encode the characteristics of the attribute (first characteristic and second characteristic). The component may include at least one of an intrinsic attribute embedding (IAE) and a contextual interaction embedding (CIE), as shown in (e) and (f) of FIG. 20.

[0133] Ideological attribute embeddings (IAEs) can be understood as embeddings that express the unique and unchanging inherent meaning of each attribute. Through these Ideological attribute embeddings, the same attribute can be learned to maintain a consistent (or identical) meaning regardless of the problem combination in which it is included. To this end, the present invention encodes analogy relationships into the embedding space so that semantic changes resulting from the addition of attributes appear identically even in different problem contexts. Through these Ideological attribute embeddings, the inherent meaning of an attribute is expressed separately from other attributes, and the model is trained to use a reusable representation across various problem combinations. In this case, the Ideological attribute embedding can be understood as including (or reflecting) a first characteristic.

[0134] Contextual interaction embeddings can be understood as embeddings that express combination-specific interaction effects occurring when multiple attributes are activated together. Based on the inherent meanings of attributes provided by implicit attribute embeddings, they can model interactions within a problem context. In other words, they can model contextual interactions added to the inherent meanings of each attribute. These contextual interaction embeddings reflect the interaction effects that appear when the same attribute is combined with other attributes, and enable the learning of expressions that vary depending on the combination, even for the same attribute. In this case, contextual interaction embeddings can be understood as including (or reflecting) a second characteristic.

[0135] More specifically, in order for the artificial intelligence model (600) to separate the learned attribute into the inherent meaning of the attribute and the interaction between the attributes, the artificial intelligence model (600) is enabled to learn an analogy for the attribute (or an analogical characteristic for the attribute).

[0136] In one embodiment, the analogy “Attribute A is for Attribute B, and Attribute C is for Attribute D (which can be expressed as A : B :: C : D)” captures the relationship between two pairs and may imply that a transformation from Attribute A to Attribute B is similar to a transformation from Attribute C to Attribute D. Let us assume that there exist two vehicle path optimization problems (e.g., Problem 1(x), Problem 2(y), etc.) in which Attribute A is not activated. In this case, variations of the vehicle path optimization problem in which Attribute A is activated can be represented as shown in (a) of FIG. 21, respectively.

[0137] At this time, based on the first characteristic examined above, the intrinsic meaning of attribute A can be expressed by an analogy as shown in Fig. 21 (b). For example, assuming x is CVRP, y is OVRP, and A is the length constraint (L), this can be expressed as “VRPL : CVRP :: OVRPL : OVRP”.

[0138] These analogous characteristics can be represented (or expressed) in the embedding space as shown in (c) of FIG. 21.

[0139] Furthermore, in the present invention, at least one pre-set loss function may be used to enable the artificial intelligence model (600) to encode the analogy relationship between attribute combinations into an embedding space. For example, the loss function may include an objective function used in contrastive learning (e.g., InfoNCE loss). The objective function may induce the artificial intelligence model (600) to learn a representation that preserves the intrinsic meaning of each attribute for the intrinsic meaning of that attribute. Additionally, the loss may also be understood as a combinational loss (see (h) in FIG. 21), and such a combinational loss may be added to a reward function (see (i) in FIG. 21). This may be represented as shown in (j) in FIG. 21.

[0140] In this regard, the calculation of combination loss can also be represented by the algorithm of Fig. 31. To describe the algorithm illustrated in Fig. 31, first, all activated attributes for each instance included in the batch are iterated through, and for each attribute (A), an attribute vector representing the meaning of that attribute is calculated. This may be a process of extracting the implicit meaning of attribute A as a difference vector in the embedding space. Subsequently, all acquired attribute vectors are stored in an attribute pool (P). Then, for each attribute vector included in the pool, other vectors with the same attribute type are selected as positive samples, and vectors corresponding to different types are formed as a negative set. Furthermore, a contrast loss in the form of InfoNCE is calculated to train the model to increase similarity between identical attributes and decrease similarity between different attributes. Finally, the average loss is returned by dividing by the total attribute pool size.

[0141] In one embodiment, let us assume that a function (see FIG. 30 (a)) representing a set of attributes activated in a problem instance (x) is given in a situation where a batch contains instances of various problem types. In this case, for any non-empty subset (see FIG. 30 (b)), an attribute vector can be extracted as shown in FIG. 30 (c). Here, a Node Embedder can be represented as in FIG. 30 (d), and a masking function that removes features of at least one attribute (e.g., Attribute A) from an instance can be represented as in FIG. 30 (e). Furthermore, in the present invention, all attribute vectors (see FIG. 30 (g)) can be collected from all instances in the batch to form an attribute pool (see FIG. 30 (f)). For the attribute vectors extracted from the above attribute pool, vectors corresponding to the same attribute type can be classified into a positive class, and vectors corresponding to different types can be classified into a negative class. One attribute vector is sampled from the positive class. In this case, when the magnitude of the negative class for a specific attribute is B, the combinatorial attribute loss can be calculated through the formula shown in (h) of FIG. 30. Additionally, the temperature parameter can be represented as shown in (i) of FIG. 30.

[0142] Here, “encoding analogical relationships between attribute combinations into an embedding space” can also be understood as encoding analogical consistency for intrinsic attribute embeddings into an embedding space. This indicates that the intrinsic meaning of a specific attribute (e.g., Attribute A) must be identifiable and consistent regardless of the base instance or other active attributes. In this case, the intrinsic meaning of the specific attribute can be represented as shown in (d) of FIG. 21. That is, in the present invention, analogical consistency can be expressed in an embedding space by distinguishing between identical attributes and different attributes.

[0143] For example, in the present invention, attribute vectors can be extracted by masking the features of the corresponding attribute in the same instance. In this case, for a specific attribute (e.g., Attribute A), a vector obtained from another instance (see FIG. 21 (e)) (see FIG. 21 (f)) can be used as a positive sample. On the other hand, vectors corresponding to different attributes (see FIG. 21 (g)) can constitute negative samples.

[0144] Meanwhile, in the present invention, the artificial intelligence model (600) may be configured to include an encoder-decoder structure. At this time, in the present invention, the encoder may be configured to further include a specific module.

[0145] For example, as illustrated in FIG. 5 (a) and (b), the artificial intelligence model (600) may be configured to include at least one of an encoder (610) and a decoder (620). In this case, the encoder (610) may be configured to further include at least one of a specific module (e.g., “ARC Module”, 611) and a node embedder (611a). In this case, the encoder (610) may be represented as in FIG. 25 (a), and the node embedder may be represented as in FIG. 25 (b).

[0146] The encoder (610) can process information of different attributes (or categories, variables, etc.). In one embodiment, the encoder (610) can process at least one of global attributes and node-specific attributes. Global attributes may include at least one of a global binary variable indicating whether a specific attribute (e.g., Open(O)) is enabled (see Fig. 25 (c)), a run time limit (see Fig. 25 (d)), and a mixed backhaul (see Fig. 25 (e)), which may represent problem-level constraints. On the other hand, node-level constraints such as linehaul demand (see Fig. 25 (f)), backhaul demand (see Fig. 25 (g)), and time windows (e.g., start time (e), end time (l), service time (s), etc.) may be associated with individual nodes.

[0147] In this case, global attribute information is integrated into the depot representation, and node attribute information (see (h) in FIG. 25) is integrated into each customer node to generate node embeddings (or initial node embeddings). These node embeddings can be represented as in FIG. 26 (a). Here, the learnable linear layer (or linear layer) can be represented as in FIG. 26 (b), and the embedding dimension can be represented as in FIG. 26 (c). Additionally, the number of global attributes and node-level attributes can be represented as in FIG. 26 (d), and the time window end value of the depot can be represented as in FIG. 26 (e).

[0148] Subsequently, the above node embedding can be input into a node embedding (611a) composed of N transformer-based blocks (e.g., NodeEmbBlock). The structure of such a node embedding (611a) can be represented as (f) in FIG. 26. Each block constituting the node embedding (611a) may be composed of at least one sublayer.

[0149] For example, the sublayer may be configured to include at least one of a first sublayer (e.g., Multi-Head Attention (MHA)) layer and a second sublayer (e.g., Feed Forward ParallelGatedMLP layer). The first sublayer captures dependencies between different locations in the input sequence, and the second sublayer may apply a non-linear transformation to the features. This can be represented as in (g) of FIG. 26. Here, the input of the l-th block can be represented as in (a) of FIG. 27, and the first sublayer can be represented as in (b) of FIG. 27. In this case, a is used as a query, and b can provide a key and a value. Additionally, RMSNorm may mean normalization.

[0150] In addition, the ParallelGatedMLP function can be represented as shown in Fig. 27 (c). Here, the element-wise multiplication can be represented as shown in Fig. 27 (d), and the SiLU can represent the Sigmoid Linear Unit (Swish) activation function. Additionally, the trainable linear layer can be represented as shown in Fig. 27 (e).

[0151] Next, a specific module (611) may be a module that performs the role of separating (or decomposing) an attribute representation into an implicit attribute embedding and a contextual interaction embedding. The specific module (611) may utilize at least one of the final output of the node embedding (611a) (i.e., the implicit attribute embedding), an attribute indicator, and a global attribute feature as input.

[0152] For example, given an input having an attribute indicator (see (f) in Fig. 27), if attribute (A) is activated, said attribute may be included in the problem. On the other hand, if the attribute is not activated, said attribute may not be included in the problem. In this case, the initial contextual interaction embedding can be represented as in (g) of Fig. 27. Here, the trainable linear layer can be represented as in (h) of Fig. 27, where Concat means concatenation of feature dimensions and LayerNorm means layer normalization.

[0153] A specific module (611) may be composed of N blocks (e.g., MixerBlock). The l-th block receives a global embedding (see (i) in FIG. 27) and an implicit attribute embedding (see (j) in FIG. 27) as inputs. The N blocks constituting this specific module (611) may be represented as shown in FIG. 28 (a). Here, the trainable linear layer may be represented as shown in FIG. 28 (b).

[0154] In addition, GlobalModule(a, b), which receives a main input (a) and a sub-input (b) and generates an output (see (c) in FIG. 28), can be calculated through the formula shown in (d) of FIG. 28.

[0155] Finally, the final output (or final output embedding) of the encoder (610) can be represented as shown in (e) of FIG. 28. In this case, the final output of the encoder (610) can be calculated by combining (or summing) the intrinsic attribute embedding and the contextual interaction embedding. This can be represented as shown in (f) of FIG. 28.

[0156] Next, the decoder (620) can output action probabilities for each node at each time point t based on the encoded node embeddings. The decoder (620) can calculate context embeddings (see (a) in FIG. 29) using the embeddings of the previously selected node and attribute feature values ​​at time point t. Here, the remaining capacity of the vehicle for line-haul and back-haul can be represented as shown in (b) in FIG. 29, and the current time, the remaining length of the partial solution, and whether there is an open path can be represented as shown in (c) in FIG. 29.

[0157] The decoder (620) can calculate the probability of selecting each node by using the context embedding as the query and the previously encoded node values ​​as the key and value. In this case, to generate a solution that satisfies the constraints, nodes that cannot be visited in each decoding process (or step) according to the activated constraints can be masked. In this case, the probability for the action can be calculated through the formula shown in (d) of FIG. 29. Here, the set of possible nodes at time t can be represented as in (e) of FIG. 29, and the clipping hyperparameter can be represented as in (f) of FIG. 29. Additionally, the dimension of the query can be represented as in (g) of FIG. 29, and the action probability can be calculated by applying a softmax to the probability value.

[0158] Meanwhile, based on the description of the artificial intelligence model (600) examined above, the structure and learning flow of the artificial intelligence model (600) according to the present invention will be briefly examined below.

[0159] Referring again to FIG. 5 (a) and (b), the encoder (610) of the artificial intelligence model (600) can receive data (21, 22, 23) containing attribute information corresponding to each of a plurality of attributes. The encoder (610) can be configured to receive the received data as input and perform the role of generating at least one embedding vector. For example, the encoder (610) can generate a final output (or final embedding, final output embedding, etc.) by using (i.e., combining) an intrinsic attribute embedding (e.g., “IAE”, 24a) corresponding to the final output of a node embedding (611a) and a contextual interaction embedding (e.g., “CIE”, 21b) corresponding to the output of a specific module (611).

[0160] Subsequently, the final output of the encoder (610) can be processed as an input to the decoder (620). Based on the final output, the decoder (620) can output a solution (25, or solution) for the data (21, 22, 23) as a prediction result. In this case, the artificial intelligence model (600) can generate first embedding data (25a) based on the solution (25).

[0161] Furthermore, the artificial intelligence model (600) can process the data (26a, 27a, 28a) from which at least one attribute has been removed from the data (21, 22, 23), respectively, to generate embedding data (e.g., second embedding data (26b), third embedding data (27b), fourth embedding data (28b), etc.). And, the artificial intelligence model (600) can learn to generate a representation vector corresponding to at least one attribute based on the difference between at least one of the embedding data (26b, 27b, 28b) and the first embedding data (25a).

[0162] As a specific example, the difference between the first embedding data and the second embedding data can be calculated by a vector difference operation. However, the present invention is not limited thereto, and the difference between the first embedding data and the second embedding data can be calculated by a distance function between vectors, a normalization operation, a projection, a linear transformation, a non-linear transformation, or a combination thereof.

[0163] In addition, a normalization operation may be performed on the result of the vector difference operation. For example, L2 normalization, scaling, or a transformation operation including a linear transformation or a non-linear activation function may be additionally performed.

[0164] In this case, it can be understood that the artificial intelligence model (600) is trained using at least one pre-set loss function (630) to perform learning based on the difference between different embedding data. More specific details regarding this will be explained later.

[0165] In another embodiment of the present invention, the artificial intelligence model (600) may also be configured with the structure shown in FIG. 6. The encoder (610) of the artificial intelligence model (600) may receive data (31, 32, 33) containing attribute information corresponding to each of a plurality of attributes. For example, the encoder (610) may generate a final output embedding by using an attribute mixing embedding (e.g., “Attribute Mixer”, 612) corresponding to the final output of a first module (e.g., “Attribute Mixer”, 612) and a node embedding (32b) corresponding to the output of a second module (e.g., “Node Embedder”, 613).

[0166] Subsequently, the final output of the encoder (610) can be processed as an input to the decoder (620). Based on the final output, the decoder (620) can output a solution (35) for the data (31, 32, 33) as a prediction result. In this case, the artificial intelligence model (600) can generate first embedding data (35a) based on the solution (35).

[0167] Furthermore, the artificial intelligence model (600) can process the data (36a, 37a, 38a) from which at least one attribute has been removed from the data (31, 32, 33), respectively, to generate embedding data (e.g., second embedding data (36b), third embedding data (37b), fourth embedding data (38b), etc.). Then, the artificial intelligence model (600) can learn to form a representation vector corresponding to at least one attribute based on the difference between at least one of the embedding data (36b, 37b, 38b) and the first embedding data (35a). In this case, the artificial intelligence model (600) can perform learning using at least one loss function (630) that has been set in advance.

[0168] Meanwhile, through the learning process described above, the artificial intelligence model (600) can be trained so that representation vectors corresponding to the same attribute included in different attribute combinations have similar representations in vector space.

[0169] For example, similarity (or similarity) can be defined based on cosine similarity. To this end, a loss function can be constructed to minimize the distance between representation vectors corresponding to the same attribute and to maximize the distance between representation vectors corresponding to different attributes. Additionally, the learning can be performed using a triplet loss or contrastive learning-based loss function.

[0170] As illustrated in FIG. 7 (a) to (c), the artificial intelligence model (600) can be trained such that, among the attributes (41a, 41b, 41c, 41d, 41e, 42a, 42b, 42c, 42d, 42e) included in the first combination in which the first attribute and the second attribute are combined, the representation vectors (41a, 41b, 41c, 41d, 41e) corresponding to the first attribute have similar representations in a vector space. Additionally, the artificial intelligence model (600) can be trained such that, among the attributes (41a, 41b, 41c, 41d, 41e, 42a, 42b, 42c, 42d, 42e) included in the first combination, the representation vectors (42a, 42b, 42c, 42d, 42e) corresponding to the second attribute have similar representations in a vector space.

[0171] In addition, the artificial intelligence model (600) can be trained such that, among the attributes (43a, 43b, 43c, 43d, 43e, 44a, 44b, 44c, 44d, 44e) included in the fourth combination in which the third attribute and the fourth attribute are combined, the representation vectors (43a, 43b, 43c, 43d, 43e) corresponding to the third attribute have similar representations in a vector space. Additionally, the artificial intelligence model (600) can be trained such that, among the attributes (43a, 43b, 43c, 43d, 43e, 44a, 44b, 44c, 44d, 44e) included in the fourth combination, the representation vectors (44a, 44b, 44c, 44d, 44e) corresponding to the fourth attribute have similar representations in a vector space.

[0172] As such, with reference to FIG. 7(c), it can be seen that the present invention forms well-separated clusters in response to variations of various problems. Additionally, it can be seen that the learned artificial intelligence model (600) according to the present invention effectively separates intrinsic components and contextual components.

[0173] Meanwhile, the attributes used in the learning or inference process of the artificial intelligence model (600) in the present invention may vary.

[0174] In this specification, the solution is described using sequential notation (see FIG. 22(a)), in which depot visits divide the sequence into K vehicle paths. For example, the solution in this specification may be represented as shown in FIG. 22(b), which may consist of K paths. Each path (see FIG. 22(c)) starts at a depot and ends at a depot (i.e., see FIG. 22(d)). All customer nodes are included in exactly one path, and the total cost may be represented as in FIG. 22(e).

[0175] Below, based on the notation examined above, we will explain the set of attributes that define variations of different problems (see (f) in Fig. 22).

[0176] In the capacity-constrained vehicle path optimization problem, customer nodes have non-negative demand (see Fig. 22 (g)), which may refer to linehaul services such as delivery. In this case, the attribute corresponding to customer node i may consist of a single demand (or demand value) (see Fig. 22 (h)). Vehicles have the same capacity (see Fig. 23 (a)). The capacity attribute requires that for each path, the sum of customer demand (see Fig. 23 (b)) must not exceed the number of customers (see Fig. 23 (c)).

[0177] In one embodiment, as illustrated in FIG. 8(e), the vehicle does not need to return to the depot (51e) after visiting all customers (54l, 54n, 54m, 54o, 54p) (i.e., after visiting the last customer). This can be represented as in FIG. 23(d), where FIG. 8(e) represents a visual example of a problem involving an Open(O) property.

[0178] In addition, unlike capacity-constrained vehicle route optimization problems where only linehaul customers exist, Backhaul (B) or Mixed Backhaul (MB) variations include backhaul customers (collection), which may mean that goods must be transported to a depot. Each route must satisfy one of two mutually exclusive properties.

[0179] In one embodiment, as illustrated in FIG. 8(a), the backhaul must start from the depot (51a), visit all linehaul customers (52a, 52b, 52c) first, and then visit the backhaul customers (53a, 53b). On the other hand, as illustrated in FIG. 8(b), the mixed backhaul may start from the depot (51b) and visit linehaul customers (52d, 52e, 52f) or backhaul customers (53c, 53d) without restriction on the order of visit. These attributes may be represented as in FIG. 23(e) and may be included in a set of global attributes (see FIG. 23(f)) as global features.

[0180] Additionally, as illustrated in FIG. 8(d), each customer node must be visited within a time interval (see FIG. 23(g)) and has a service time (see FIG. 23(f)). Thus, customer characteristics can be defined as shown in FIG. 24(a). If a vehicle arrives earlier than the earliest possible arrival time (see FIG. 24(b)), it must wait. The service takes the service time, after which the vehicle moves to the next node. In this case, the time windows (55a, 55b, 55c, 55d, 55e, 55f) of the depot (51c) can be represented as in FIG. 24(c), where T represents the total time limit. When the Time Window (55a, 55b, 55c, 55d, 55e, 55f) attribute is not enabled, the Time Window (TW) for all customers (54g, 54h, 54i, 54j, 54k) can be set as shown in (d) of FIG. 24.

[0181] Furthermore, as illustrated in FIG. 8(c), the total cost of each route from the depot (51c) to all customers (54a, 54b, 54c, 54e, 54f) (see FIG. 24(e)) cannot exceed a limit value (i.e., see FIG. 24(f)). If this attribute is not enabled, the limit value can be set to infinity (see FIG. 24(g)). A global feature (see FIG. 24(h)) includes a limit value, which can be represented as in FIG. 24(i). FIG. 8(c) shows a visual example of a problem including the attribute of a total duration limit (L).

[0182] Meanwhile, the control unit (700) can perform the role of controlling the overall operation of the artificial intelligence inference system (1000) related to the present invention. The control unit (700) can process signals, data, information, etc. that are input or output through the components of the artificial intelligence inference system (1000) described above, or perform a series of data processing to provide or process appropriate information and functions to the user. The control unit (700) can be physically implemented by the processor described above.

[0183] Meanwhile, as described above, the present invention provides an artificial intelligence reasoning system (1000) that solves various problems through reasoning using artificial intelligence. The artificial intelligence reasoning system (1000) according to the present invention may include an artificial intelligence model (600) that can be universally applied to various problems and usefully utilized to solve various problems. The present invention aims to efficiently solve various problems by flexibly responding to various problems using such an artificial intelligence model (600).

[0184] Hereinafter, the learning and inference method of artificial intelligence according to the present invention will be examined in more detail together with the attached drawings. FIG. 9a is a flowchart for explaining the inference method of artificial intelligence according to the present invention, and FIG. 9b is a flowchart for explaining the learning method of artificial intelligence according to the present invention. Furthermore, FIG. 10, FIG. 11, FIG. 12, FIG. 13, FIG. 14, FIG. 15, FIG. 16, FIG. 17, FIG. 18, FIG. 19, FIG. 20, FIG. 21, FIG. 22, FIG. 23, FIG. 24, FIG. 25, FIG. 26, FIG. 27, FIG. 28, FIG. 29, FIG. 30, and FIG. 31 are conceptual diagrams and formulas for explaining the learning and inference method of artificial intelligence according to the present invention.

[0185] Meanwhile, as illustrated in FIG. 9b, the learning method of artificial intelligence according to the present invention may include the steps of receiving data containing attribute information corresponding to each of a plurality of attributes (S951), inputting the data into a prediction model to generate first embedding data (S953), inputting data from which at least one attribute has been removed into a prediction model to generate second embedding data (S955), and training the prediction model to form a representation vector corresponding to at least one attribute based on the difference between the first embedding data and the second embedding data (S957).

[0186] In one embodiment, a plurality of attributes may include constraints or operational conditions that must be considered in various problems. In this case, the plurality of attributes may include at least one of resource constraints (e.g., vehicle capacity limit (e.g., (Capacity), load limit, etc.), time constraints (e.g., time window, operation time limit, etc.), route structure constraints (e.g., return status (e.g., Open), visit order constraint (e.g., Backhaul), etc.) and operation range constraints (e.g., total travel distance limit, total operation time limit (e.g., Duration Limit), etc.). However, the examples included in the plurality of attributes in the present invention are not necessarily limited to those mentioned, and various additional attributes may be included depending on the domain of the problem or the domain of the data.

[0187] The data processing unit (500) can receive (or collect) data (original data) containing attribute information corresponding to each of a plurality of attributes from at least one of various sources (e.g., a website, a database (DB), an application programming interface (API) linked to an artificial intelligence inference system (1000), a server linked to an artificial intelligence inference system (1000), a central server, an external server, cloud storage, a user terminal, a large dataset, etc.).

[0188] For example, the data processing unit (500) can receive data including attribute information corresponding to a first attribute (e.g., resource constraint), attribute information corresponding to a second attribute (e.g., time constraint), and attribute information corresponding to a third attribute (e.g., driving range constraint).

[0189] Subsequently, the data processing unit (500) can process the received data as input to an artificial intelligence model (600, or prediction model) to generate first embedding data. Here, the first embedding data may be formed to include an intrinsic attribute embedding representing the unique meaning of each attribute, and a contextual interaction embedding representing the interaction that occurs as multiple attributes are activated together. That is, the first embedding data is generated by reflecting the combination of multiple attributes included in the received data, and can be understood as an expression that integrally represents the structural information of the problem instance and the activated attribute information.

[0190] In this case, the intrinsic attribute embedding may be expressed as “first embedding component” or “first embedding” in this specification, and the contextual interaction embedding may be expressed as “second embedding component” or “second embedding”.

[0191] Specifically, when the encoder (610) of the artificial intelligence model (600) receives data containing attribute information corresponding to each of a plurality of attributes, it can generate a final output (or final embedding, final output embedding, etc.) based on the data.

[0192] As seen above, the node embedding (611a) constituting the encoder (610) can output a first embedding component representing a unique meaning corresponding to each of the multiple attributes.

[0193] Additionally, a specific module (611) constituting the encoder (610) may be configured to generate a second embedding component by utilizing at least one of the output of the node embeddinger (611a) (e.g., a first embedding component), an attribute indicator, and a global attribute feature as input. At this time, the specific module (611) may be a module that performs the role of separating the attribute representation into a first embedding component and a second embedding component. That is, the artificial intelligence model (600) may be trained through the specific module (611) so that a first embedding component representing a unique meaning corresponding to each of a plurality of attributes and a second embedding component corresponding to the interaction of the plurality of attributes are separated and formed.

[0194] Afterwards, the encoder (610) can combine the separated first embedding component and the second embedding component to generate a final input, and generate first embedding data (1310) based on the final input (see (a) of FIG. 10).

[0195] More specifically, the encoder (610) can generate a final output by combining the first embedding component and the second embedding component, and generate the first embedding data (1310) by inputting the final output to the decoder (620). That is, the first embedding data (1310) can also be understood as being formed by the combination of the first embedding component and the second embedding component.

[0196] Furthermore, in the present invention, data in which at least one attribute among a plurality of attributes included in the received data has been removed can be input into an artificial intelligence model (600) to generate second embedding data.

[0197] Here, the process of removing at least one attribute may include a masking process for removing at least one attribute among a plurality of attributes that is to be removed.

[0198] Specifically, the system (1000) can perform masking to remove at least one attribute among a plurality of attributes included in the received data, thereby generating data in which at least one attribute is removed.

[0199] For example, as illustrated in FIG. 10 (a) and (b), the system (1000) can generate data in which the first attribute (1311) is removed by performing masking to remove the first attribute (1311) among a plurality of attributes included in the data (e.g., a first attribute (e.g., resource constraint, 1311), a second attribute (e.g., time constraint, 1312), a third attribute (e.g., driving range constraint, 1313), etc.).

[0200] As another example, as illustrated in FIG. 10 (a) and (c), the system (1000) can generate data (1330) from which the second attribute (1312) has been removed by performing masking to remove the second attribute (1312) among a plurality of attributes (1311, 1312, 1313) included in the data.

[0201] As another example, as illustrated in FIG. 10 (a) and (d), the system (1000) can generate data (1340) from which the third attribute (1313) has been removed by performing masking to remove the third attribute (1313) among a plurality of attributes (1311, 1312, 1313) included in the data.

[0202] Subsequently, the system (1000) can generate second embedding data by inputting data in which at least one attribute has been removed through masking into an artificial intelligence model (600). For example, the system (1000) can generate second embedding data (1320) by inputting data in which the first attribute (1311) has been removed into an artificial intelligence model (600).

[0203] In this regard, the first embedding data (1310) and the second embedding data (1320) may be generated depending on whether at least one attribute is included for the same data.

[0204] More specifically, the first embedding data (1310) and the second embedding data (1320) may be generated based on the same original data. The data includes multiple attribute information, and the first embedding data (1310) is generated through an artificial intelligence model (600) while including all attribute information corresponding to each of the multiple attributes (1311, 1312, 1313). That is, the first embedding data (1310) may be the embedding result of the same data that includes all of the multiple attributes (or attribute information corresponding to each of the multiple attributes).

[0205] On the other hand, the second embedding data (1320) can be generated by inputting it into the artificial intelligence model (600) in a state where at least one attribute information (1311) is removed or disabled from the same original data. That is, the second embedding data (1320) may be the embedding result of the same data from which at least one attribute has been removed.

[0206] As such, the first embedding data (1310) and the second embedding data (1320) may be embedding results generated by differing only in whether a specific attribute is included for the same data sample.

[0207] Additionally, the system (1000) can generate third embedding data (1330) by inputting data with the second attribute (1312) removed into an artificial intelligence model (600), and generate fourth embedding data (1340) by inputting data with the third attribute (1313) removed into an artificial intelligence model (600).

[0208] Meanwhile, in the present invention, an artificial intelligence model (600) can be trained to form a representation vector corresponding to at least one attribute based on the difference between the first embedding data (1310) and the second embedding data (1320).

[0209] In the present invention, a representation vector may refer to a vector that represents (or indicates) the amount of semantic change induced in the vector space (or embedding space) of an artificial intelligence model (600) by including a specific attribute in the data. That is, the representation vector corresponding to a specific attribute may be calculated as the vector difference between the embedding data when the specific attribute is present and the embedding data when the specific attribute is removed. This may represent the unique semantic contribution or characteristic change induced by the specific attribute in the embedding space.

[0210] First, as illustrated in FIG. 11 (a) and FIG. 12 (a), the system (1000) can calculate (or operate) the difference (or loss) between the first embedding data (1310) and the second embedding data (1320) using at least one preset loss function, and generate a representation vector (1321) corresponding to the first attribute based on the calculated difference. This representation vector (1321) corresponding to the first attribute can be understood as a vector representing the effect that the removed at least one attribute (i.e., the first attribute (1311)) has on the overall representation of the data containing a plurality of attributes (1311, 1312, 1313). In this case, the representation vector (1321) corresponding to the first attribute may include at least one attribute vector (1321a, 1321b, 1321c) corresponding to the first attribute.

[0211] Additionally, as illustrated in FIG. 11(b) and FIG. 12(b), the system (1000) can calculate the difference between the first embedding data (1310) and the third embedding data (1330) using the loss function and generate a representation vector (1331) corresponding to the second attribute based on the calculated difference. This representation vector (1331) corresponding to the second attribute can be understood as a vector representing the effect of the removed at least one attribute (i.e., the second attribute (1312)) on the overall representation of the data containing a plurality of attributes (1311, 1312, 1313). In this case, the representation vector (1331) corresponding to the second attribute may include at least one attribute vector (1331a, 1331b, 1331c) corresponding to the second attribute.

[0212] Next, as illustrated in FIG. 11 (c) and FIG. 12 (c), the system (1000) can use the loss function to calculate the difference between the first embedding data (1310) and the fourth embedding data (1340) and generate a representation vector (1341) corresponding to a third attribute based on the calculated difference. This representation vector (1341) corresponding to a third attribute can be understood as a vector representing the effect of the removed at least one attribute (i.e., the third attribute (1313)) on the overall representation of the data containing a plurality of attributes (1311, 1312, 1313). In this case, the representation vector (1341) corresponding to a third attribute may include at least one attribute vector (1341a, 1341b, 1341c) corresponding to the third attribute.

[0213] Furthermore, the artificial intelligence model (600) can be trained to form a representation vector corresponding to at least one attribute based on the difference between different embedding data using at least one loss function that is set in advance.

[0214] For example, as illustrated in FIG. 13, an artificial intelligence model (600) can be trained so that representation vectors (1321a, 1321b, 1321c, 1321d) corresponding to a first attribute have similar representations in a vector space (i.e., become closer or have a higher similarity). In this case, the artificial intelligence model (600) can be trained so that representation vectors corresponding to at least one attribute (e.g., a second attribute, a third attribute, etc.) different from the representation vectors (1321a, 1321b, 1321c, 1321d) corresponding to the first attribute do not have similar representations (i.e., become further apart or have a lower similarity).

[0215] In another example, the artificial intelligence model (600) can be trained so that the representation vectors (1331a, 1331b, 1331c) corresponding to the second attribute have similar representations in the vector space (i.e., become close to each other). In this case, the artificial intelligence model (600) can be trained so that the representation vectors corresponding to at least one attribute (e.g., first attribute, third attribute, etc.) different from the representation vectors (1331a, 1331b, 1331c) corresponding to the second attribute do not have similar representations to each other.

[0216] As another example, the artificial intelligence model (600) can be trained so that the representation vectors (1341a, 1341b, 1341c) corresponding to the third attribute have similar representations in a vector space. In this case, the artificial intelligence model (600) can be trained so that the representation vectors corresponding to at least one attribute (e.g., first attribute, second attribute, etc.) different from the representation vectors (1341a, 1341b, 1341c) corresponding to the third attribute do not have similar representations.

[0217] In this way, when the same attribute is included for different data or problems, the representation vector is trained to have a direction with similar directionality and magnitude in the embedding space. Accordingly, the artificial intelligence model (600) is trained so that the representation vector corresponding to a specific attribute maintains a consistent meaning even if the attribute is included in various contexts. Through this, the trained artificial intelligence model (600) can distinguish and express the inherent meaning of the attribute and the interaction effect between the combination of multiple attributes.

[0218] Meanwhile, the artificial intelligence inference system (1000) according to the present invention includes an artificial intelligence model (600) trained using the learning method described above, and can solve various problems through inference using the trained artificial intelligence model (600). In this regard, as illustrated in FIG. 9a, the artificial intelligence inference method according to the present invention may include the step of receiving a prediction request (S910), the step of inputting the prediction request into a prediction model trained such that a representation vector corresponding to a plurality of attributes is reflected in an embedding (S920), and the step of generating a prediction result based on at least one specific attribute forming the prediction request by the prediction model (S930). However, the order of the steps for processing the prediction request is not limited to one embodiment, and in some implementations, the attribute analysis step may be performed before receiving the prediction request, or a separate post-processing step may be added after inputting the prediction request.

[0219] For example, the system (1000) may receive at least one data (or input data) corresponding to (or included in) a prediction request. The prediction request may include multiple attribute information. In this case, the data may be data containing attributes (or attribute information) corresponding to each of the multiple attributes. Alternatively, the data may include a problem containing (or composed of) at least one attribute. Such a prediction request may be processed as input to a learned artificial intelligence model (600).

[0220] For example, the prediction request may consist of data in the form of a vector, matrix, or tensor. Additionally, the prediction request may include graph structure data, and the graph structure data may be represented in a form including nodes and edges.

[0221] As seen above, the trained artificial intelligence model (600, or prediction model) may be a model trained such that representation vectors have similar representations in a vector space. For example, the artificial intelligence model may include a prediction model trained to reflect the change in embeddings based on whether an attribute is included.

[0222] In this case, the representation vector corresponding to multiple attributes may be a vector corresponding to the same attribute included in different attribute combinations. More specifically, the learned artificial intelligence model (600) is a model learned for various attribute combinations in which multiple attributes are combined, and each attribute may be included in the input data either alone or in a combined form with other attributes. At this time, even if the same specific attribute is included in different attribute combinations, the representation vector corresponding to the specific attribute may be formed to have a consistent meaning in vector space.

[0223] For example, let us assume that a first attribute combination includes a first attribute and a second attribute, and a second attribute combination includes a first attribute and a third attribute. The representation vector corresponding to the first attribute can be formed as a vector having a similar representation when based on whether it is included in the first attribute combination or the second attribute combination. That is, the representation vector reflects the unique meaning corresponding to the same attribute regardless of the overall context of the attribute combination.

[0224] This learned artificial intelligence model (600) may be a model learned by aligning the embedding differences between different combinations of attributes that include the same attribute, or a model learned by applying constraints so that the amount of change in embeddings based on whether the same attribute is included is similar to each other.

[0225] That is, representation vectors corresponding to the same attribute can be placed in a vector space to have stable direction and magnitude regardless of whether they are combined with other attributes. Based on this, the artificial intelligence model (600) trained can i) reuse the meaning of a specific attribute independently of the attribute combination, and even if a new attribute combination not used during the training process is input, it can generate a prediction result based on the representation vectors corresponding to the same attribute, and knowledge sharing between multiple attribute combinations is possible, thereby improving generalization performance.

[0226] Meanwhile, the representation vector corresponding to the multiple attributes may be generated during the learning process of the artificial intelligence model (600) based on the difference between the first embedding data and the second embedding data.

[0227] In this case, the first embedding data and the second embedding data can each be generated by data distinguished based on whether at least one attribute is included. For example, the second embedding data may be embedding data generated by inputting data from which at least one specific attribute has been removed into an artificial intelligence model (600). Since more specific details regarding this have been explained above, they will be explained only briefly to avoid duplication of explanation.

[0228] The above representation vector can be combined as part of an embedding to form an embedding component corresponding to a specific attribute. More specifically, the trained artificial intelligence model (600) can convert an input prediction request (or data included in the prediction request) into a vector-shaped embedding and generate a prediction result based on the embedding.

[0229] For example, the above prediction result can be generated by inputting the above embedding into a decoder. The decoder can output a final prediction result based on the first embedding component corresponding to the specific attribute. However, the present invention is not limited thereto, and the decoder can output a final prediction result by receiving as input the first embedding component corresponding to the specific attribute or the combined result of the first embedding component and the second embedding component.

[0230] In this case, the above-mentioned representation vector may be included as part of the element vectors constituting the embedding. For example, the embedding may be formed by combining multiple representation vectors corresponding to each of multiple attributes, or by combining multiple attribute-specific representation vectors with vectors corresponding to other input information. Here, “combination” can be implemented in various ways, such as simple sum, weighted sum, concatenation, synthesis after linear transformation, or integration through neural network layers.

[0231] The above-mentioned embedding may include a first embedding component representing a unique meaning corresponding to each of a plurality of attributes, and a second embedding component related to the interaction of the plurality of attributes. The first embedding component and the second embedding component do not necessarily have to be generated by physically separated modules, but may be implemented in a subspace decomposition method within a single integrated network. In this case, the embedding (i.e., embedding data) may be formed by the combination of the first embedding component and the second embedding component. That is, the learned artificial intelligence model (600) may generate the embedding by combining the first embedding component and the second embedding component. The method of combining the first embedding component and the second embedding component may include at least one of simple summation, concatenation, weighted summation, attention-based consolidation, gating structure, or other neural network-based consolidation methods.

[0232] Meanwhile, the trained artificial intelligence model (600) can generate a prediction result based on at least one specific attribute that forms a prediction request.

[0233] In this case, the prediction result can be generated based on a first embedding component corresponding to a specific attribute. For example, a trained artificial intelligence model (600) can generate an embedding by processing data included in a prediction request and generate a prediction result based on the embedding. At this time, the embedding may include a first embedding component representing a unique meaning corresponding to each attribute and a second embedding component related to the interaction of multiple attributes.

[0234] The first embedding component is a component learned to independently represent a meaning corresponding to a specific attribute, and can reflect whether the specific attribute is included and the semantic characteristics of the attribute. In this case, the prediction result can be generated based on the first embedding component corresponding to the specific attribute. For example, i) the learned artificial intelligence model (600) may extract or emphasize the first embedding component corresponding to the specific attribute among the embeddings and reflect it in the prediction operation, ii) calculate the final output value by assigning weights to the first embedding component, or iii) select different prediction paths by forming a branch structure based on the first embedding component.

[0235] Since this first embedding component can be formed to maintain a consistent meaning as a vector corresponding to the same attribute included in different attribute combinations, even when a new attribute combination not used in the learning process is input, a prediction result can be generated based on the first embedding component corresponding to the specific attribute.

[0236] Meanwhile, the learned artificial intelligence model (600) can generate a prediction result based on a representation vector corresponding to at least one specific attribute even when an attribute combination including at least one specific attribute is not used in the learning process.

[0237] That is, the learned artificial intelligence model (600) can generate a prediction result even when an attribute combination that was not used during the learning process is input.

[0238] In one embodiment, as illustrated in FIG. 14, the system (1000) may input an attribute combination (1400) including a plurality of attributes (1311, 1313, 1411) to a learned artificial intelligence model (600). At this time, even if the attribute combination (1400) is an attribute combination that is not used in the learning process of the artificial intelligence model (600), the learned artificial intelligence model (600) may generate a prediction result (1450) based on representation vectors (1321a, 1321d, 1321c, 1321d) corresponding to at least one specific attribute (1311) among the attributes included in the attribute combination (1400).

[0239] In this case, referring to FIG. 15, the representation vectors (1421a, 1421b, 1421c, 1421d) corresponding to the new attributes included in the new attribute combination may have similar characteristics (i.e., representations) to the representation vectors (1321a, 1321d, 1321c, 1321d) corresponding to a specific attribute (1311). That is, they may have similar representations in vector space. Here, having similar characteristics may mean that the inherent meaning of the attribute (e.g., the first embedding component) is similar.

[0240] That is, the learned artificial intelligence model (600) can also be understood as generating a prediction result based on the first embedding component corresponding to the specific attribute (1311), even when a new combination of attributes not used in the learning process is input.

[0241] Meanwhile, the expression vector can be formed to represent a meaning corresponding to the specific attribute independently of other attribute combinations that include the specific attribute.

[0242] More specifically, the learned artificial intelligence model (600) can be trained so that the representation vector corresponding to the specific attribute has a consistent meaning in vector space even when the same specific attribute is included in different combinations of attributes.

[0243] To this end, as seen above, during the learning process of the artificial intelligence model (600), a constraint may be imposed so that the amount of change in embedding depending on whether the specific attribute is included is similar even between different combinations of attributes. Accordingly, the representation vector corresponding to the specific attribute can be formed so that its meaning is not distorted by interaction with other attributes.

[0244] Meanwhile, the artificial intelligence model (600) trained using the learning method according to the present invention can be universally applied to various problems and utilized effectively.

[0245] In one embodiment, the artificial intelligence model (600) may be utilized as a model specialized in at least one domain. Here, the domain may include a domain related to combinatorial optimization problems.

[0246] In this case, the trained artificial intelligence model (600) may be a model trained to output a solution to a combinatorial optimization problem as the prediction result. However, the present invention is not limited to combinatorial optimization problems and can be applied to any data processing environment where a combination of multiple attributes changes dynamically. For example, it can be applied to various technical fields such as communication network resource allocation, manufacturing process scheduling, cloud computing resource distribution, and medical diagnostic variable combination analysis.

[0247] Combinatorial optimization problems may include at least one of resource allocation problems, path determination problems, or scheduling problems. Such combinatorial optimization problems can be defined as problems of seeking a solution that optimizes an objective function under limited resources or conditions.

[0248] For example, a resource allocation problem may involve allocating multiple resources to multiple targets while satisfying constraints such as capacity, time, or cost. Additionally, a path determination problem may involve determining a path to visit multiple nodes while generating a path that satisfies constraints such as distance, time, capacity, and visit order. Furthermore, a scheduling problem may involve placing multiple tasks along a time axis while generating a schedule that satisfies precedence, deadlines, and resource constraints.

[0249] In this case, the plurality of attributes may include attributes that define constraints. That is, each attribute can be utilized as information indicating whether a specific constraint is enabled or the specific details of the constraint. For example, the plurality of attributes may include at least one of an attribute corresponding to a capacity limit, an attribute corresponding to a time limit, an attribute corresponding to a path length limit, or an attribute corresponding to the availability of a specific resource.

[0250] Meanwhile, the artificial intelligence inference system (1000) according to the present invention can be applied to various environments and utilized effectively.

[0251] In one embodiment, as illustrated in FIGS. 16 and 17, the inference system (1000) of artificial intelligence according to the present invention can be applied to a simulation environment (or a simulation platform, a simulation system, etc.) and utilized effectively. In this case, a user terminal (10) may be provided with a user interface capable of configuring a manufacturing process environment and applying various control algorithms.

[0252] For example, as illustrated in FIG. 16, among the plurality of regions included in the screen (1600) of the user interface, the first region (1610) can define the grid structure of the process layout space by adjusting the number of rows and columns, and can create a process layout through a button (e.g., “Create”). Accordingly, the user can define the spatial structure of the process environment and set the basic physical structure of the production line.

[0253] Additionally, among the multiple areas included in the screen (1600) of the user interface, the second area (1620) is configured to allow the selection of multiple components constituting the process, such as an assembly station, a conveyor, a robot, and a work node. In this case, the selected components can be placed in an interactive visual editor (1601). The interactive visual editor (1601) provides a grid-based process layout screen and can be configured to allow the user to place components in a drag-and-drop manner and visually design the process flow. This editor can go beyond simple visualization functions and can visually represent animations according to state changes, bottleneck locations, and changes in the workflow.

[0254] Additionally, among the multiple areas included in the screen (1600) of the user interface, the third area (1630) is configured to allow setting dynamic parameters such as conveyor speed, feed interval, working time, and processing capacity. These parameters can be used as environmental variables that affect state transitions and compensation calculations when running a simulation.

[0255] As another example, as illustrated in FIG. 17, among the plurality of regions included in the screen (1600) of the user interface, the fourth region (1640) is configured to allow the selection of a control method. In this case, the control method may include at least one of heuristic-based control, reinforcement learning-based control (RL), and large-scale language model-based control (LLM). The policy decision method may be changed according to the selected control method.

[0256] Additionally, among the multiple areas included in the screen (1600) of the user interface, the fifth area (1650) is configured to allow selection of one of the following: execution speed control, Play, Pause, Step, or Stop. This allows visual confirmation of the reinforcement learning process or the policy inference process.

[0257] Additionally, the animation view (1601) can display workpiece movement, robot operation status, bottleneck occurrence, waiting status, etc. in an animation form.

[0258] Furthermore, among the real-time monitoring areas (1602, 1603, 1604), the first monitoring area (1602) can display changes in production throughput over time in real time. Through this, changes in production efficiency due to policy changes can be analyzed. In addition, the second monitoring area (1603) visualizes the robot's status (Idle, Moving, Interacting, Charging, etc.) in real time, which can be used for resource utilization analysis. Also, the third monitoring area (1604) can determine whether policy performance has improved by displaying cumulative rewards over time when reinforcement learning-based control is selected. This can be used as an indicator to check learning stability and convergence.

[0259] As described above, according to the artificial intelligence inference method and system of the present invention, the prediction model according to the present invention is trained to form attribute-specific representation vectors based on the difference between first embedding data and second embedding data, thereby enabling the precise extraction of the unique meaning of each attribute by separating it from other attribute combinations. Accordingly, the meaning of the same attribute is consistently maintained even between different problem combinations, and stable prediction results can be generated even for new attribute combinations that have not been learned. In other words, the present invention can improve the prediction accuracy and generalization performance of the prediction model regarding combination optimization problems.

[0260] In addition, according to the artificial intelligence inference method and system of the present invention, by separating and configuring a first embedding component representing the intrinsic meaning of each attribute and a second embedding component corresponding to the interaction of multiple attributes, the prediction model is enabled to learn by structurally decomposing the intrinsic meaning and contextual interaction of attributes. Through this, the prediction model is trained to maintain similar representations in a vector space even when the same attribute is included in different attribute combinations, and can consistently preserve analogy relationships between attributes in the embedding space. As a result, knowledge sharing between problem types can be efficiently achieved, and high performance can be achieved with a single model for various modified problems.

[0261] Furthermore, according to the artificial intelligence inference method and system of the present invention, a prediction model is trained such that embedding transformations resulting from the addition or removal of specific attributes have similar directionality even in different problem contexts, thereby enabling the formation of analogous relationships between attributes in the embedding space. Through this, the prediction model can secure generalization performance capable of generating accurate prediction results even for problems involving new combinations that have not been learned. Therefore, the present invention can flexibly respond to combinations of constraints that continuously change in actual industrial environments.

[0262] Furthermore, the artificial intelligence inference method and system according to the present invention can be applied to combinatorial optimization problems involving the combination of multiple constraints. By decomposing each constraint into attribute units and vectorizing their representations, it is possible to derive stable and precise solutions even in complex constraint environments. In other words, the present invention can simultaneously improve the accuracy of problem solving and generalization performance in real-world logistics, transportation, and scheduling environments where various constraints change dynamically.

[0263] Furthermore, according to the artificial intelligence inference method and system of the present invention, by structuring attribute representations into a reusable form, rapid adaptation with a small amount of data is possible even when new attributes or combinations of new attributes are added. Through this, the present invention can respond to new problem types solely through efficient parameter adjustment without the need to retrain the entire model from scratch, thereby reducing training costs and time resources.

[0264] Meanwhile, the present invention described above can be implemented based on a quantum computer. The present invention implemented based on a quantum computer may include a qubit-based quantum processor and quantum memory, and may include software and hardware interfaces optimized for quantum computation.

[0265] Quantum processors in quantum computers utilize qubits to efficiently process complex operations through parallel computation, quantum entanglement, and quantum superposition, which cannot be performed by the binary bits of classical computers. Quantum processors process data using quantum gates and can provide exponential speed improvements for specific problems.

[0266] Meanwhile, the present invention described above can be implemented as a program that is executed by one or more processes on a computer and can be stored on a computer-readable medium (or recording medium).

[0267] Furthermore, the present invention described above can be implemented as computer-readable code or instructions on a medium on which a program is recorded. That is, the present invention can be provided in the form of a program.

[0268] Meanwhile, computer-readable media include all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include HDD (Hard Disk Drive), SSD (Solid State Disk), SSD (Silicon Disk Drive), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0269] Furthermore, the computer-readable medium may be a server or cloud storage that includes a storage and is accessible to an electronic device via communication. In this case, the computer may download the program according to the present invention from the server or cloud storage via wired or wireless communication.

[0270] A computer program may reach the system (1000) through various suitable transmission mechanisms. The transmission mechanism may be, for example, a computer-readable storage medium, a computer program product, a memory device, a recording medium such as a CD-ROM or DVD, or a product that tangibly embodies the computer program. The transmission mechanism may be a signal configured to reliably transmit the computer program through air or an electrical connection. The system (1000) may propagate or transmit the computer program as a computer data signal.

[0271] Furthermore, references to 'computer-readable storage media,' 'computer program products,' 'computer programs embodied in a tangible form,' etc., or to 'controller,' 'computer,' 'processor,' etc., should be understood to include not only computers with various architectures such as single / t-processor architectures and sequential (Von Neumann) / sequence architectures, but also specialized circuits such as Field-Programmable Gate Arrays (FPGAs), Application Specific Circuits (ASICs), signal processing units, and other devices. References to computer programs, instructions, code, etc., should be understood to include software for programmable processors or firmware, such as programmable content for hardware devices, whether it is instructions for a processor or configuration settings for a fixed-function device, gate array, or programmable logic device.

[0272] Furthermore, in the present invention, the computer described above is an electronic device equipped with a processor, namely a CPU (Central Processing Unit), and no special limitations are placed on its type.

[0273] Meanwhile, the above detailed description should not be interpreted restrictively in all respects but should be considered exemplary. The scope of the invention shall be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the invention are included within the scope of the invention.

Claims

1. Regarding methods performed by a computer, Step of receiving a prediction request; A step of inputting the prediction request into a prediction model trained such that representation vectors corresponding to multiple attributes are reflected in the embedding; and A step of generating a prediction result based on at least one specific attribute forming the prediction request by the above prediction model; An artificial intelligence reasoning method characterized by including 2. In Paragraph 1, The representation vector corresponding to the above plurality of attributes is, It is generated during the learning process of the prediction model based on the difference between the first embedding data and the second embedding data, and An artificial intelligence inference method characterized in that the first embedding data and the second embedding data are each generated by data distinguished according to whether or not at least one attribute is included.

3. In Paragraph 2, An artificial intelligence inference method characterized in that the second embedding data is embedding data generated by inputting the data from which at least one attribute has been removed into the prediction model.

4. In Paragraph 1, The above prediction model is, An artificial intelligence inference method characterized by the fact that the above-mentioned representation vectors are learned to have similar representations in a vector space.

5. In Paragraph 4, The above representation vector is, An artificial intelligence inference method characterized by being a vector corresponding to the same attribute included in a combination of different attributes.

6. In Paragraph 4, The above representation vector is, An artificial intelligence inference method characterized by being combined as part of the above embedding to form an embedding component corresponding to the above specific attribute.

7. In Paragraph 1, The above embedding is, A first embedding component representing a unique meaning corresponding to each attribute, and An artificial intelligence inference method characterized by including a second embedding component related to the interaction of multiple attributes.

8. In Paragraph 7, An artificial intelligence inference method characterized in that the above embedding is formed by the combination of the above first embedding component and the above second embedding component.

9. In Paragraph 7, The above prediction results are, An artificial intelligence inference method characterized by being generated based on the first embedding component corresponding to the specific attribute mentioned above.

10. In Paragraph 1, An artificial intelligence inference method characterized by the fact that even when a combination of attributes including at least one specific attribute is not used in the learning process of the prediction model, the prediction model generates a prediction result based on a representation vector corresponding to the at least one specific attribute.

11. In Paragraph 10, The above representation vector is, An artificial intelligence reasoning method characterized by being formed to represent a meaning corresponding to the specific attribute independently of other attribute combinations that include the specific attribute.

12. In Paragraph 1, The above prediction model is trained to output a solution to a combinatorial optimization problem as the above prediction result, and The above combinational optimization problem includes at least one of resource placement, path determination, or scheduling problems combined with multiple constraints, and An artificial intelligence reasoning method characterized in that the above plurality of attributes include attributes defining the above constraints.

13. In a method performed by a computer, A step of receiving data including attribute information corresponding to each of a plurality of attributes; A step of generating first embedding data by inputting the above data into a prediction model; A step of generating second embedding data by inputting data from which at least one attribute has been removed from the above data into the prediction model; and A step of training the prediction model so that a representation vector corresponding to at least one attribute is formed based on the difference between the first embedding data and the second embedding data; A learning method for artificial intelligence characterized by including 14. In Paragraph 13, A learning method for artificial intelligence characterized in that the first embedding data and the second embedding data are generated based on whether the at least one attribute is included for the same data.

15. In Paragraph 13, A learning method for artificial intelligence characterized in that the above second embedding data is generated by inputting the data from which at least one attribute has been removed into the prediction model.

16. In Paragraph 13, A learning method for artificial intelligence characterized by training the prediction model such that the representation vectors corresponding to the same attribute included in different attribute combinations have similar representations in a vector space.

17. In Paragraph 13, A learning method for artificial intelligence characterized in that the above prediction model is learned such that a first embedding component representing a unique meaning corresponding to each attribute and a second embedding component corresponding to the interaction of multiple attributes are formed separately.

18. In Paragraph 17, A learning method for artificial intelligence characterized by learning to form the first embedding data by combining the first embedding component and the second embedding component.

19. Processor; and It includes memory connected to the above processor, and The above memory stores a prediction model trained such that representation vectors corresponding to multiple attributes are reflected in the embedding, and The above processor is, Receive a prediction request, Input the above prediction request into the above prediction model, An artificial intelligence inference system characterized by being formed to generate a prediction result based on at least one specific attribute among a plurality of attributes constituting the above prediction request.

20. A program that is executed by one or more processes in an electronic device and stored on a computer-readable recording medium, The above program is, Step of receiving a prediction request; A step of inputting the prediction request into a prediction model trained such that representation vectors corresponding to multiple attributes are reflected in the embedding; and A program characterized by including instructions that perform the step of generating a prediction result based on at least one specific attribute forming the prediction request by the prediction model.

Images