Artificial intelligence server

Abstract

An artificial intelligence server is disclosed. The artificial intelligence server includes a communication unit that communicates with an electronic device, and a processor that updates the classification layer by providing training data for classification and labeling data for classification to an artificial intelligence model that includes a feature extraction layer that extracts a feature vector and a classification layer that classifies input data using the feature vector, and transmits the updated classification layer to the electronic device.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence server capable of updating and providing a classification layer among the feature extraction layer and classification layer constituting an artificial intelligence model. Background Technology

[0002] Artificial intelligence is a field of computer science and information technology that studies methods to enable computers to perform thinking, learning, and self-development capable of human intelligence, and refers to making computers mimic intelligent human behavior.

[0003] Furthermore, artificial intelligence does not exist in isolation but is closely related, directly and indirectly, to many other fields of computer science. Particularly in the modern era, there are very active attempts to introduce AI elements into various sectors of information technology and utilize them to solve problems within those fields.

[0004] Meanwhile, technologies that utilize artificial intelligence to perceive and learn surrounding situations, provide information desired by the user in the desired format, or perform actions or functions desired by the user are being actively researched.

[0005] And electronic devices that provide these various operations and functions can be referred to as artificial intelligence devices.

[0006] A model trained through machine learning cannot go beyond the learning range during inference.

[0007] Taking face recognition as an example, if training is performed using data predominantly consisting of Caucasians, it is evident that the inference accuracy regarding Asians will inevitably decrease due to the discrepancy with the training data.

[0008] In addition, if the entire AI model is retrained using training data focused on Asians to ensure inference accuracy for Asians, there is a problem in that the same level of cost as when it was trained first is incurred again.

[0009] Furthermore, general-purpose models trained on Caucasians and Asians may not be able to deliver performance that meets specific user requirements. For example, if a user employs an AI model solely to classify Caucasians, a model optimized for Caucasians may perform better than a general-purpose model. The problem to be solved

[0010] The present invention aims to solve the aforementioned problems, and the objective of the present invention is to provide an artificial intelligence server capable of updating and providing a classification layer among the feature extraction layer and classification layer constituting an artificial intelligence model. means of solving the problem

[0011] An artificial intelligence server according to one embodiment of the present invention includes a communication unit that communicates with an electronic device, and a processor that updates the classification layer by providing training data for classification and labeling data for classification to an artificial intelligence model including a feature extraction layer that extracts a feature vector and a classification layer that classifies input data using the feature vector, and transmits the updated classification layer to the electronic device. Effects of the invention

[0012] There is a need to update artificial intelligence models. Examples include cases where classification for new categories is required (such as updating an existing AI model that classified apples, oranges, and tomatoes to additionally classify watermelons) or cases where performance for categories frequently used by users is improved (such as updating an existing AI model that classified Black, White, and Asian people to classify Asians more effectively).

[0013] Furthermore, according to the present invention, since only the classification layer is updated and distributed, there is an advantage in that the amount of training data, training time, and the amount of distributed data can be reduced compared to a method in which both the feature extraction layer and the classification layer are updated and distributed. Brief explanation of the drawing

[0014] FIG. 1 shows an AI device (100) according to one embodiment of the present invention. FIG. 2 shows an AI server (200) according to one embodiment of the present invention. FIG. 3 shows an AI system (1) according to one embodiment of the present invention. Figure 4 is a diagram illustrating the operation method of an artificial intelligence server. Figure 5 is a diagram illustrating a CNN among neural networks. Figure 6 is a diagram illustrating a method for creating an artificial intelligence model by training a CNN. Figure 7 is a diagram illustrating the deployment of a generated artificial intelligence model. Figure 8 is a diagram illustrating how to update a classification hierarchy. Figure 9 is a diagram illustrating the distribution of classification layers. FIG. 10 is a diagram illustrating a method for generating multiple classification layers and providing them to an electronic device. Figure 11 is a diagram illustrating a method for replacing classification layers. Figure 12 is a diagram illustrating a method for providing a classification hierarchy of categories desired by the user. Figure 13 is a diagram illustrating another method of providing a classification hierarchy of categories desired by the user. Specific details for implementing the invention

[0015] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components regardless of drawing symbols will be assigned the same reference number, 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 inherently possess distinct meanings or roles. Furthermore, in describing 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.

[0016] 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.

[0017] 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.

[0018] Artificial Intelligence (AI)

[0019] Artificial intelligence refers to the field of researching artificial intelligence or the methodologies to create it, while machine learning refers to the field of researching methodologies to define and solve various problems addressed within the field of artificial intelligence. Machine learning is also defined as an algorithm that improves performance on a task through continuous experience.

[0020] An Artificial Neural Network (ANN) is a model used in machine learning that can refer to any model capable of problem-solving, composed of artificial neurons (nodes) that form a network through the connection of synapses. An artificial neural network can be defined by connection patterns between neurons in different layers, a learning process that updates model parameters, and an activation function that generates output values.

[0021] An artificial neural network may include an input layer, an output layer, and optionally one or more hidden layers. Each layer may include one or more neurons, and the artificial neural network may include synapses connecting the neurons. In an artificial neural network, each neuron may output a function value of an activation function for input signals, weights, and biases input through the synapses.

[0022] Model parameters refer to parameters determined through learning, including synaptic connection weights and neuron biases. Hyperparameters, on the other hand, refer to parameters that must be set prior to training in a machine learning algorithm, including the learning rate, number of iterations, mini-batch size, and initialization function.

[0023] The objective of training an artificial neural network can be viewed as determining model parameters that minimize the loss function. The loss function can be used as an indicator to determine optimal model parameters during the training process of an artificial neural network.

[0024] Machine learning can be classified into supervised learning, unsupervised learning, and reinforcement learning depending on the learning method.

[0025] Supervised learning refers to a method of training an artificial neural network with labels provided for the training data; a label can refer to the correct answer (or result) that the neural network must infer when the training data is input. Unsupervised learning refers to a method of training an artificial neural network without labels provided for the training data. Reinforcement learning refers to a learning method in which an agent defined within an environment is trained to select an action or sequence of actions that maximizes the cumulative reward in each state.

[0026] Machine learning implemented using a Deep Neural Network (DNN) that includes multiple hidden layers among artificial neural networks is also called Deep Learning, and Deep Learning is a part of Machine Learning. Hereinafter, Machine Learning is used in a sense that includes Deep Learning.

[0027] Robot

[0028] A robot can refer to a machine that automatically processes or operates a given task based on its own capabilities. In particular, a robot that has the ability to perceive its environment, make decisions on its own, and perform actions can be called an intelligent robot.

[0029] Robots can be classified into industrial, medical, household, military, etc., depending on their purpose of use or field.

[0030] A robot can perform various physical movements, such as moving robot joints, by being equipped with a drive unit including actuators or motors. Additionally, a mobile robot may include wheels, brakes, propellers, etc., in the drive unit, enabling it to drive on the ground or fly in the air.

[0031] Self-Driving

[0032] Autonomous driving refers to technology that drives itself, and an autonomous vehicle refers to a vehicle that drives without user intervention or with minimal user intervention.

[0033] For example, autonomous driving can include technologies such as maintaining the driving lane, automatically adjusting speed like adaptive cruise control, automatically driving along a set route, and automatically setting a route and driving once a destination is set.

[0034] Vehicles encompass vehicles equipped solely with internal combustion engines, hybrid vehicles equipped with both internal combustion engines and electric motors, and electric vehicles equipped solely with electric motors, and may include not only automobiles but also trains, motorcycles, etc.

[0035] In this case, an autonomous vehicle can be viewed as a robot with autonomous driving capabilities.

[0036] Extended Reality (XR)

[0037] Extended Reality is a collective term for Virtual Reality (VR), Augmented Reality (AR), and Mixed Reality (MR). VR technology provides real-world objects or backgrounds solely as CG images, AR technology provides virtual CG images superimposed on real-world images, and MR technology is a computer graphics technology that mixes and combines virtual objects with the real world.

[0038] MR technology is similar to AR technology in that it displays real-world objects and virtual objects together. However, there is a difference in that while virtual objects in AR technology are used to complement real-world objects, virtual objects and real-world objects are used as equals in MR technology.

[0039] XR technology can be applied to HMDs (Head-Mount Displays), HUDs (Head-Up Displays), mobile phones, tablet PCs, laptops, desktops, TVs, digital signage, etc., and devices to which XR technology is applied can be called XR devices.

[0040] FIG. 1 shows an AI device (100) according to one embodiment of the present invention.

[0041] The AI ​​device (100) can be implemented as a stationary device or a mobile device, such as a TV, projector, mobile phone, smartphone, desktop computer, laptop, digital broadcasting terminal, PDA (personal digital assistants), PMP (portable multimedia player), navigation, tablet PC, wearable device, set-top box (STB), DMB receiver, radio, washing machine, refrigerator, desktop computer, digital signage, robot, vehicle, etc.

[0042] Referring to FIG. 1, the terminal (100) may include a communication unit (110), an input unit (120), a learning processor (130), a sensing unit (140), an output unit (150), a memory (170), and a processor (180), etc.

[0043] The communication unit (110) can transmit and receive data with external devices, such as other AI devices (100a to 100e) or an AI server (200), using wired or wireless communication technology. For example, the communication unit (110) can transmit and receive sensor information, user input, learning models, control signals, etc., with external devices.

[0044] At this time, the communication technology used by the communication unit (110) includes GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), LTE (Long Term Evolution), 5G, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity), Bluetooth (Bluetooth™), RFID (Radio Frequency Identification), Infrared Data Association (IrDA), ZigBee, NFC (Near Field Communication), etc.

[0045] The input unit (120) can acquire various types of data.

[0046] At this time, the input unit (120) may include a camera for inputting a video signal, a microphone for receiving an audio signal, a user input unit for receiving information from a user, etc. Here, the camera or microphone may be treated as a sensor, and the signal obtained from the camera or microphone may be referred to as sensing data or sensor information.

[0047] The input unit (120) can obtain training data for model training and input data to be used when obtaining an output using a training model. The input unit (120) may also obtain unprocessed input data, in which case the processor (180) or the learning processor (130) can extract input feature points as a preprocessing step for the input data.

[0048] The learning processor (130) can train a model composed of an artificial neural network using training data. Here, the trained artificial neural network may be referred to as a learning model. The learning model can be used to infer a result value for new input data other than the training data, and the inferred value can be used as a basis for judgment to perform an action.

[0049] At this time, the learning processor (130) can perform AI processing together with the learning processor (240) of the AI ​​server (200).

[0050] At this time, the learning processor (130) may include memory integrated into or implemented in the AI ​​device (100). Alternatively, the learning processor (130) may be implemented using memory (170), external memory directly coupled to the AI ​​device (100), or memory maintained in an external device.

[0051] The sensing unit (140) can obtain at least one of internal information of the AI ​​device (100), surrounding environment information of the AI ​​device (100), and user information using various sensors.

[0052] At this time, the sensors included in the sensing unit (140) include a proximity sensor, an illuminance sensor, an accelerometer, a magnetic sensor, a gyroscope, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, a light sensor, a microphone, a lidar, a radar, etc.

[0053] The output unit (150) can generate output related to sight, hearing, or touch.

[0054] At this time, the output unit (150) may include a display unit that outputs visual information, a speaker that outputs auditory information, a haptic module that outputs tactile information, etc.

[0055] The memory (170) can store data that supports various functions of the AI ​​device (100). For example, the memory (170) can store input data, training data, training models, training history, etc. obtained from the input unit (120).

[0056] The processor (180) can determine at least one executable action of the AI ​​device (100) based on information determined or generated using a data analysis algorithm or a machine learning algorithm. The processor (180) can perform the determined action by controlling the components of the AI ​​device (100).

[0057] To this end, the processor (180) can request, search, receive, or utilize data from the learning processor (130) or memory (170), and can control the components of the AI ​​device (100) to execute a predicted operation or a preferred operation among the at least one executable operation.

[0058] At this time, if the processor (180) requires the connection of an external device to perform a determined operation, it can generate a control signal to control the external device and transmit the generated control signal to the external device.

[0059] The processor (180) can obtain intent information regarding user input and determine the user's requirements based on the obtained intent information.

[0060] At this time, the processor (180) can obtain intent information corresponding to the user input by using at least one of a Speech To Text (STT) engine for converting voice input into a string or a Natural Language Processing (NLP) engine for obtaining intent information of natural language.

[0061] At this time, at least one of the STT engine or NLP engine may be composed of an artificial neural network in which at least a portion is learned according to a machine learning algorithm. Also, at least one of the STT engine or NLP engine may be learned by a learning processor (130), learned by a learning processor (240) of an AI server (200), or learned through distributed processing thereof.

[0062] The processor (180) may collect history information, including the operation details of the AI ​​device (100) or user feedback regarding the operation, and store it in memory (170) or a learning processor (130), or transmit it to an external device such as an AI server (200). The collected history information may be used to update a learning model.

[0063] The processor (180) can control at least some of the components of the AI ​​device (100) to run an application stored in memory (170). Furthermore, the processor (180) can operate two or more of the components included in the AI ​​device (100) in combination with each other to run the application.

[0065] FIG. 2 shows an AI server (200) according to one embodiment of the present invention.

[0066] Referring to FIG. 2, the AI ​​server (200) may refer to a device that trains an artificial neural network using a machine learning algorithm or uses a trained artificial neural network. Here, the AI ​​server (200) may be composed of multiple servers to perform distributed processing and may be defined as a 5G network. At this time, the AI ​​server (200) may be included as part of the configuration of the AI ​​device (100) and may perform at least part of the AI ​​processing together.

[0067] The AI ​​server (200) may include a communication unit (210), memory (230), a learning processor (240), and a processor (260), etc.

[0068] The communication unit (210) can transmit and receive data with external devices such as AI devices (100).

[0069] The memory (230) may include a model storage unit (231). The model storage unit (231) may store a model (or artificial neural network, 231a) that is being learned or has been learned through a learning processor (240).

[0070] The learning processor (240) can train the artificial neural network (231a) using training data. The training model may be used while mounted on the AI ​​server (200) of the artificial neural network, or it may be used while mounted on an external device such as an AI device (100).

[0071] The learning model may be implemented in hardware, software, or a combination of hardware and software. If part or all of the learning model is implemented in software, one or more instructions constituting the learning model may be stored in memory (230).

[0072] The processor (260) can use a learning model to infer a result value for new input data and generate a response or control command based on the inferred result value.

[0073] FIG. 3 shows an AI system (1) according to one embodiment of the present invention.

[0074] Referring to FIG. 3, the AI ​​system (1) is connected to a cloud network (10) at least one of an AI server (200), a robot (100a), an autonomous vehicle (100b), an XR device (100c), a smartphone (100d), or a home appliance (100e). Here, the robot (100a), the autonomous vehicle (100b), the XR device (100c), the smartphone (100d), or the home appliance (100e) to which AI technology is applied may be referred to as AI devices (100a to 100e).

[0075] A cloud network (10) may mean a network that constitutes part of a cloud computing infrastructure or exists within a cloud computing infrastructure. Here, the cloud network (10) may be configured using a 3G network, a 4G or LTE (Long Term Evolution) network or a 5G network, etc.

[0076] That is, each device (100a to 100e, 200) constituting the AI ​​system (1) can be connected to each other through a cloud network (10). In particular, each device (100a to 100e, 200) may communicate with each other through a base station, but may also communicate directly with each other without going through a base station.

[0077] The AI ​​server (200) may include a server that performs AI processing and a server that performs operations on big data.

[0078] The AI ​​server (200) is connected via a cloud network (10) to at least one of the AI ​​devices constituting the AI ​​system (1), such as a robot (100a), an autonomous vehicle (100b), an XR device (100c), a smartphone (100d), or a home appliance (100e), and can assist in at least some of the AI ​​processing of the connected AI devices (100a to 100e).

[0079] At this time, the AI ​​server (200) can train an artificial neural network according to a machine learning algorithm on behalf of the AI ​​devices (100a to 100e), and can directly store the training model or transmit it to the AI ​​devices (100a to 100e).

[0080] At this time, the AI ​​server (200) receives input data from the AI ​​devices (100a to 100e), infers a result value for the received input data using a learning model, and generates a response or control command based on the inferred result value and transmits it to the AI ​​devices (100a to 100e).

[0081] Alternatively, the AI ​​device (100a to 100e) may use a direct learning model to infer a result value for input data and generate a response or control command based on the inferred result value.

[0082] Hereinafter, various embodiments of AI devices (100a to 100e) to which the above-described technology is applied will be described. Here, the AI ​​devices (100a to 100e) illustrated in FIG. 3 can be seen as specific embodiments of the AI ​​device (100) illustrated in FIG. 1a.

[0083] <AI+로봇>

[0084] The robot (100a) can be implemented as a guide robot, transport robot, cleaning robot, wearable robot, entertainment robot, pet robot, unmanned flying robot, etc. by applying AI technology.

[0085] The robot (100a) may include a robot control module for controlling operation, and the robot control module may mean a software module or a chip that implements the same in hardware.

[0086] The robot (100a) can use sensor information obtained from various types of sensors to obtain state information of the robot (100a), detect (recognize) surrounding environment and objects, generate map data, determine movement path and driving plan, determine response to user interaction, or determine action.

[0087] Here, the robot (100a) can use sensor information obtained from at least one sensor among lidar, radar, and camera to determine a movement path and driving plan.

[0088] The robot (100a) can perform the above-mentioned actions using a learning model composed of at least one artificial neural network. For example, the robot (100a) can recognize the surrounding environment and objects using the learning model, and can determine actions using the recognized surrounding environment information or object information. Here, the learning model may be learned directly by the robot (100a) or learned from an external device such as an AI server (200).

[0089] At this time, the robot (100a) may perform an operation by generating a result using a direct learning model, but it may also perform an operation by transmitting sensor information to an external device such as an AI server (200) and receiving the result generated accordingly.

[0090] The robot (100a) can determine a movement path and a driving plan using at least one of map data, object information detected from sensor information, or object information obtained from an external device, and control a driving unit to drive the robot (100a) according to the determined movement path and driving plan.

[0091] Map data may include object identification information for various objects placed in the space where the robot (100a) moves. For example, map data may include object identification information for fixed objects such as walls and doors, and movable objects such as flowerpots and desks. In addition, the object identification information may include names, types, distances, locations, etc.

[0092] Additionally, the robot (100a) can perform actions or drive by controlling the drive unit based on the user's control / interaction. At this time, the robot (100a) can acquire intention information of interaction based on the user's actions or voice utterances, and can perform actions by determining a response based on the acquired intention information.

[0093] <AI+자율주행>

[0094] The autonomous vehicle (100b) can be implemented as a mobile robot, vehicle, unmanned aerial vehicle, etc. by applying AI technology.

[0095] The autonomous vehicle (100b) may include an autonomous driving control module for controlling autonomous driving functions, and the autonomous driving control module may refer to a software module or a chip that implements the same in hardware. The autonomous driving control module may be included internally as a component of the autonomous vehicle (100b), but may also be configured and connected as separate hardware externally to the autonomous vehicle (100b).

[0096] The autonomous vehicle (100b) can use sensor information obtained from various types of sensors to obtain state information of the autonomous vehicle (100b), detect (recognize) surrounding environment and objects, generate map data, determine a travel path and driving plan, or determine an action.

[0097] Here, the autonomous vehicle (100b) can use sensor information obtained from at least one sensor among lidar, radar, and camera, just like the robot (100a), to determine a travel path and a driving plan.

[0098] In particular, the autonomous vehicle (100b) can recognize environments or objects in areas where the field of view is obscured or in areas beyond a certain distance by receiving sensor information from external devices, or by receiving information directly recognized from external devices.

[0099] The autonomous vehicle (100b) can perform the above-mentioned operations using a learning model composed of at least one artificial neural network. For example, the autonomous vehicle (100b) can recognize surrounding environments and objects using the learning model, and can determine a driving path using the recognized surrounding environment information or object information. Here, the learning model may be learned directly in the autonomous vehicle (100b) or learned from an external device such as an AI server (200).

[0100] At this time, the autonomous vehicle (100b) may perform operations by generating results using a direct learning model, but may also perform operations by transmitting sensor information to an external device such as an AI server (200) and receiving the results generated accordingly.

[0101] The autonomous vehicle (100b) can determine a movement path and a driving plan using at least one of map data, object information detected from sensor information or object information obtained from an external device, and control a driving unit to drive the autonomous vehicle (100b) according to the determined movement path and driving plan.

[0102] Map data may include object identification information for various objects placed in a space (e.g., a road) where the autonomous vehicle (100b) is driving. For example, the map data may include object identification information for fixed objects such as streetlights, rocks, and buildings, and movable objects such as vehicles and pedestrians. In addition, the object identification information may include names, types, distances, locations, etc.

[0103] Additionally, the autonomous vehicle (100b) can perform operations or drive by controlling the drive unit based on the user's control / interaction. At this time, the autonomous vehicle (100b) can acquire intention information of the interaction based on the user's actions or voice utterances, and can perform operations by determining a response based on the acquired intention information.

[0104] <AI+XR>

[0105] The XR device (100c) can be implemented as a Head-Mount Display (HMD), a Head-Up Display (HUD) equipped in a vehicle, a television, a mobile phone, a smartphone, a computer, a wearable device, a home appliance, digital signage, a vehicle, a stationary robot, or a mobile robot by applying AI technology.

[0106] The XR device (100c) can obtain information about surrounding space or real objects by analyzing 3D point cloud data or image data obtained through various sensors or from an external device to generate position data and attribute data for 3D points, and can render and output an XR object to be output. For example, the XR device (100c) can output an XR object containing additional information about a recognized object by associating it with the recognized object.

[0107] The XR device (100c) can perform the above-mentioned operations using a learning model composed of at least one artificial neural network. For example, the XR device (100c) can recognize real-world objects in 3D point cloud data or image data using the learning model and can provide information corresponding to the recognized real-world objects. Here, the learning model may be learned directly by the XR device (100c) or learned from an external device such as an AI server (200).

[0108] At this time, the XR device (100c) may perform an operation by generating a result using a direct learning model, but it may also perform an operation by transmitting sensor information to an external device such as an AI server (200) and receiving the result generated accordingly.

[0109] <AI+로봇+자율주행>

[0110] The robot (100a) can be implemented as a guide robot, transport robot, cleaning robot, wearable robot, entertainment robot, pet robot, unmanned flying robot, etc. by applying AI technology and autonomous driving technology.

[0111] A robot (100a) equipped with AI technology and autonomous driving technology may refer to the robot itself having autonomous driving capabilities, or a robot (100a) that interacts with an autonomous driving vehicle (100b).

[0112] A robot (100a) with autonomous driving capabilities can be collectively referred to as a device that moves on its own along a given path without user control, or moves by determining its own path.

[0113] A robot (100a) and an autonomous vehicle (100b) having autonomous driving capabilities may use a common sensing method to determine one or more of a travel path or a driving plan. For example, a robot (100a) and an autonomous vehicle (100b) having autonomous driving capabilities may determine one or more of a travel path or a driving plan by using information sensed through a lidar, radar, or camera.

[0114] A robot (100a) interacting with an autonomous vehicle (100b) exists separately from the autonomous vehicle (100b) and can perform actions linked to the autonomous driving function inside the autonomous vehicle (100b) or linked to a user riding in the autonomous vehicle (100b).

[0115] At this time, the robot (100a) interacting with the autonomous vehicle (100b) can control or assist the autonomous driving function of the autonomous vehicle (100b) by acquiring sensor information on behalf of the autonomous vehicle (100b) and providing it to the autonomous vehicle (100b), or by acquiring sensor information and generating surrounding environment information or object information and providing it to the autonomous vehicle (100b).

[0116] Alternatively, a robot (100a) interacting with an autonomous vehicle (100b) may monitor a user riding in the autonomous vehicle (100b) or control the functions of the autonomous vehicle (100b) through interaction with the user. For example, if the robot (100a) determines that the driver is drowsy, it may activate the autonomous driving function of the autonomous vehicle (100b) or assist in controlling the drive unit of the autonomous vehicle (100b). Here, the functions of the autonomous vehicle (100b) controlled by the robot (100a) may include not only the autonomous driving function but also functions provided by a navigation system or an audio system equipped inside the autonomous vehicle (100b).

[0117] Alternatively, a robot (100a) interacting with an autonomous vehicle (100b) may provide information to or assist functions to the autonomous vehicle (100b) from outside the autonomous vehicle (100b). For example, the robot (100a) may provide traffic information, such as signal information, to the autonomous vehicle (100b), such as a smart traffic light, or may interact with the autonomous vehicle (100b) to automatically connect an electric charger to the charging port, such as an automatic electric charger for an electric vehicle.

[0118] <AI+로봇+XR>

[0119] The robot (100a) can be implemented as a guide robot, transport robot, cleaning robot, wearable robot, entertainment robot, pet robot, unmanned flying robot, drone, etc. by applying AI technology and XR technology.

[0120] A robot (100a) to which XR technology is applied may refer to a robot that is the subject of control / interaction within an XR image. In this case, the robot (100a) is distinguished from the XR device (100c) and can be interconnected with it.

[0121] When a robot (100a) that is the target of control / interaction within an XR image acquires sensor information from sensors including a camera, the robot (100a) or the XR device (100c) can generate an XR image based on the sensor information, and the XR device (100c) can output the generated XR image. Furthermore, the robot (100a) can operate based on a control signal input through the XR device (100c) or user interaction.

[0122] For example, the user can view an XR image corresponding to the viewpoint of the remotely linked robot (100a) through an external device such as an XR device (100c), and through interaction, can adjust the autonomous driving path of the robot (100a), control its movement or driving, or check information about surrounding objects.

[0123] <AI+자율주행+XR>

[0124] The autonomous vehicle (100b) can be implemented as a mobile robot, vehicle, unmanned aerial vehicle, etc. by applying AI technology and XR technology.

[0125] An autonomous vehicle (100b) equipped with XR technology may refer to an autonomous vehicle equipped with means for providing XR images, or an autonomous vehicle that is the subject of control / interaction within the XR images. In particular, an autonomous vehicle (100b) that is the subject of control / interaction within the XR images may be distinguished from an XR device (100c) and may be interconnected with it.

[0126] An autonomous vehicle (100b) equipped with means for providing XR images can acquire sensor information from sensors including cameras and output an XR image generated based on the acquired sensor information. For example, the autonomous vehicle (100b) can provide an XR object corresponding to a real object or an object in the screen to the occupant by providing an XR image by outputting an XR image with a HUD.

[0127] At this time, when the XR object is displayed on the HUD, at least a portion of the XR object may be displayed so as to overlap with the actual object to which the occupant's gaze is directed. On the other hand, when the XR object is displayed on a display provided inside the autonomous vehicle (100b), at least a portion of the XR object may be displayed so as to overlap with an object on the screen. For example, the autonomous vehicle (100b) may display XR objects corresponding to objects such as lanes, other vehicles, traffic lights, traffic signs, motorcycles, pedestrians, buildings, etc.

[0128] When an autonomous vehicle (100b) that is the subject of control / interaction within an XR image acquires sensor information from sensors including a camera, the autonomous vehicle (100b) or the XR device (100c) can generate an XR image based on the sensor information, and the XR device (100c) can output the generated XR image. Furthermore, the autonomous vehicle (100b) can operate based on control signals input through an external device such as the XR device (100c) or user interaction.

[0129] Meanwhile, the term AI device (100) can be used interchangeably with the term artificial intelligence server (100).

[0130] Figure 4 is a diagram illustrating the operation method of an artificial intelligence server.

[0131] Referring to FIG. 4, the method of operation of an artificial intelligence server may include the step of creating an artificial intelligence model by training a neural network using training data for artificial intelligence training and labeling data for artificial intelligence training (S410), the step of updating a classification layer by training the artificial intelligence model (S430), and the step of transmitting the updated classification layer to an electronic device (S450).

[0132] In the following, neural networks are explained using the example of a CNN (Convolutional Neural Network). However, this is not limited to this, and neural networks may include various networks such as DNN (Deep Neural Network), RNN (Recurrent Neural Network), BRDNN (Bidirectional Recurrent Deep Neural Network), and MLP (Multilayer Perceptron).

[0133] Figure 5 is a diagram illustrating a CNN among neural networks.

[0134] Figure 5 illustrates a general form of CNN.

[0135] A CNN is a type of Deep Neural Network (DNN) that can consist of an input layer, one or more convolutional layers, one or more pooling layers, one or more fully connected layers, and an output layer. Here, the pooling layer can also be referred to as subsampling.

[0136] CNNs have a structure suitable for learning 2D data, so they can be widely used in various application fields such as object classification and object detection within images.

[0137] A CNN may include a feature extraction layer that extracts feature vectors and a classification layer that classifies input data using the extracted feature vectors.

[0138] Here, the feature extraction layer may include one or more convolutional layers and one or more pooling layers.

[0139] Meanwhile, the feature extraction layer may include multiple convolutional layers and multiple pooling layers, in which case the convolutional layers and pooling layers may be arranged alternately.

[0140] And the feature extraction layer can receive input data, extract a feature vector from the received input data, and output it.

[0141] Specifically, the feature extraction layer can extract a feature vector from input data using alternating convolution layers and integration layers, an activation function such as ReLU (Rectified Linear Unit), and pre-set parameters (at least one of weights and biases).

[0142] Meanwhile, the classification layer may include one or more fully connected layers, and in some cases, the classification layer may further include an output layer.

[0143] The classification layer receives feature vectors and can classify input data using the received feature vectors.

[0144] Specifically, the classification layer can classify input data into one of a plurality of classes using one or more fully connected layers and pre-configured parameters (at least one of weights and biases).

[0145] Meanwhile, the following description uses an example of a CNN but is not limited thereto, and the present invention can be applied to an artificial intelligence model (or neural network) comprising a feature extraction layer that extracts feature vectors and a classification layer that classifies input data using feature vectors.

[0146] Figure 6 is a diagram illustrating a method for creating an artificial intelligence model by training a CNN.

[0147] The processor can train a CNN using general-purpose training data and general-purpose labeling data.

[0148] Here, general-purpose training data refers to training data used to create an artificial intelligence model that is universally applicable to input data of various categories. Additionally, general-purpose labeling data refers to the data provided to a CNN as the ground truth values ​​for the general-purpose training data in supervised learning.

[0149] For example, when a CNN is trained to generate a face recognition model, the general training data may include images of faces of Caucasians, Black people, and Asians. Additionally, the general labeling data may be Caucasian if the general training data consists of images of Caucasians, Black if the general training data consists of images of Black people, and Asian if the general training data consists of images of Asians.

[0150] Meanwhile, the CNN (620) receives general-purpose training data and can extract feature vectors of the general-purpose training data according to parameters set in the feature extraction layer (625). Additionally, the CNN can output an estimate according to parameters set in the classification layer (630). Here, the estimate may refer to the result of the classification layer classifying the feature vector into one of a plurality of classes.

[0151] During the process of training the neural network, the processor can adjust the parameters of the feature extraction layer (625) (at least one of weights and biases) and the parameters of the classification layer (630) (at least one of weights and biases) so that the error between the neural network's estimated value and the general-purpose labeling data is reduced.

[0152] Specifically, the processor can change the parameters of the artificial intelligence model (620) in a direction that minimizes the cost (the error between the neural network's estimated value and the general-purpose labeling data) based on a cost function. Here, a softmax-based cost function may be used as the cost function, but is not limited thereto.

[0153] In addition, the processor can update not only the parameters of the classification layer (630) but also the parameters of the feature extraction layer (625) based on the backpropagation algorithm.

[0154] That is, the processor can propagate the cost (or error) in the reverse direction of the processing direction of the CNN, first update the parameters of the classification layer (630) based on the cost (or error), and then update the parameters of the feature extraction layer (625). In this case, a gradient descent algorithm may be used to find the parameters that minimize the cost (or error).

[0155] Meanwhile, when the CNN (620) achieves a certain level of accuracy through iterative training using general-purpose training data and general-purpose labeling data, the processor can complete the training.

[0156] Figure 7 is a diagram illustrating the deployment of a generated artificial intelligence model.

[0157] A CNN that exhibits a certain level of accuracy through iterative training using general-purpose training data and general-purpose labeling data can be referred to as an artificial intelligence model (720).

[0158] Meanwhile, the parameter of the feature extraction layer (725) in the state where training is completed can be referred to as the first parameter. Additionally, the parameter of the classification layer (730) in the state where training is completed can be referred to as the second-1 parameter.

[0159] That is, the generated artificial intelligence model (720) may include a 'feature extraction layer (725) that extracts a feature vector using a first parameter' and a 'classification layer (730) that classifies input data using a feature vector using a second-first parameter'.

[0160] As previously explained, general-purpose training data refers to training data used to create artificial intelligence models that are universally applicable to input data of various categories.

[0161] And the artificial intelligence model (720) trained using general-purpose training data can be distributed and used on multiple electronic devices (710).

[0162] Specifically, the processor can transmit the artificial intelligence model (720) to the electronic device (710). The electronic device (710) can download the artificial intelligence model (720) and use it for recognition.

[0163] Meanwhile, the usage environment of electronic device users can be very diverse.

[0164] For example, users primarily use input data from a specific category, but data from that specific category may not be included in the general training data or may be included in only a small amount. For instance, an AI model designed to classify types of round fruits was trained using general training data that includes images of apples, oranges, and tomatoes, but a user might use the AI ​​model to classify watermelons. In this case, the performance of the AI ​​model decreases.

[0165] As another example, users primarily use input data from a specific category, but data from that specific category is used as training data at an equivalent level to data from other categories. For instance, an AI model designed to classify Black, White, and Asian people is trained using general-purpose training data that includes images of Black, White, and Asian individuals; however, a user might use this AI model to classify Asians while excluding Black and White people. In this case, an AI model that classifies Black, White, and Asian people universally may perform worse than an AI model specialized for classifying Asians.

[0166] However, if the entire feature extraction layer and classification layer are retrained, there is a problem in that the same level of cost as when it was first trained is incurred again.

[0167] However, in the present invention, an artificial intelligence model optimized for an individual can be provided by updating and distributing only the classification layer.

[0168] Figure 8 is a diagram illustrating how to update a classification hierarchy.

[0169] The processor can train the generated artificial intelligence model (or the artificial intelligence model distributed to the electronic device) (720).

[0170] Specifically, the processor can train the artificial intelligence model (720) by providing training data for classification and labeling data for classification to the artificial intelligence model (720).

[0171] Here, classification training data may refer to training data intended to improve the performance of an artificial intelligence model on input data of a specific category. Additionally, classification training data may refer to training data belonging to a specific category.

[0172] For example, classification training data may be used to improve the performance of a face recognition model on input data of a specific category (Asians). Therefore, classification training data may include a large number of images of Asian faces. However, it is not limited to this, and images of Caucasians and African Americans may also be used as classification training data along with images of Asian faces.

[0173] As another example, the training data for classification may be training data intended to improve the performance of a fruit classification model for input data of a specific category (watermelon). Therefore, the training data for classification may include multiple images of watermelons. However, it is not limited to this, and images of other fruits along with images of watermelons may also be used as training data for classification.

[0174] Meanwhile, classification labeling data may be data provided to an artificial intelligence model as the correct answer value of the classification training data in supervised learning.

[0175] For example, if classification training data (multiple watermelon images) corresponding to a specific category is provided to an artificial intelligence model, the classification labeling data may be watermelon.

[0176] Meanwhile, the processor can update the classification layer by training the artificial intelligence model, which includes a feature extraction layer and a classification layer, by providing training data for classification and labeling data for classification to the artificial intelligence model.

[0177] Specifically, the artificial intelligence model (720) receives training data for classification and can extract feature vectors of the training data for classification according to parameters set in the feature extraction layer (725). Here, the parameters set in the feature extraction layer (725) may be the first parameters described above.

[0178] Additionally, the artificial intelligence model (720) can output an estimated value according to the parameters set in the classification layer. Here, the estimated value may refer to the result of the classification layer classifying a feature vector into one of a plurality of classes. Here, the parameters set in the classification layer may be the 2-1 parameters described above. That is, the classification layer generated by training the CNN using the general-purpose training data can be reused.

[0179] The processor can obtain a classification layer (830) having a second-2 parameter different from the second-1 parameter by providing training data for classification and labeling data for classification to the artificial intelligence model (720) and training the artificial intelligence model.

[0180] Specifically, during the process of training the artificial intelligence model (720), the processor can adjust the parameters of the classification layer (at least one of weights and biases) so that the error between the estimated value of the artificial intelligence model (720) and the labeling data for classification becomes smaller.

[0181] More specifically, the processor can modify the parameters of the classification layer based on a cost function in a direction that minimizes the cost (the error between the neural network's estimate and the general-purpose labeling data). Here, a Triplet Loss method can be used as the cost function, but is not limited thereto.

[0182] It should be noted here that, unlike the training process of the CNN mentioned earlier, the parameters of the feature extraction layer (725) are not changed.

[0183] Specifically, in the training process of the CNN mentioned earlier, not only the parameters of the classification layer but also the parameters of the feature extraction layer were updated based on the backpropagation algorithm.

[0184] However, during the training process of the artificial intelligence model, the processor can adjust the parameters of the classification layer so that the error between the estimated value of the artificial intelligence model (720) and the labeling data for classification is reduced, and the parameters of the feature extraction layer (725) can be maintained as they were before.

[0185] Meanwhile, if the artificial intelligence model (720) shows an accuracy of a certain level or higher through repeated training using classification training data and classification labeling data, the processor can complete the training of the artificial intelligence model (720).

[0186] Meanwhile, since the cost (or error) was not backpropagated to the feature extraction layer, the parameters of the feature extraction layer in the state where training is completed can remain the same as before the training of the artificial intelligence model (720) even after the artificial intelligence model is trained.

[0187] However, due to the training of the artificial intelligence model, the parameter of the classification layer changed from the 2-1 parameter to the 2-2 parameter, which is different from the 2-1 parameter.

[0188] That is, the trained artificial intelligence model may include a feature extraction layer (725) having the same parameters as before training (first parameters) and a classification layer (830) having different parameters from before training (second-2).

[0189] Meanwhile, as explained above, the parameter set in the classification layer when starting the training of the artificial intelligence model may be the 2-1 parameter. However, it is not limited to this, and the parameter set in the classification layer when starting the training of the artificial intelligence model may be the initial parameter.

[0190] Specifically, the processor can replace the classification layer having the 2-1 parameter with the classification layer having the initial parameter, and train the artificial intelligence model by providing training data for classification and labeling data for classification to the artificial intelligence model.

[0191] In this case, the processor can adjust the parameters of the classification layer (at least one of weights and biases) so that the error between the estimated value of the artificial intelligence model (720) and the labeling data for classification becomes smaller.

[0192] And when the artificial intelligence model (720) shows an accuracy of a certain level or higher through repeated training using classification training data and classification labeling data, the processor can complete the training of the artificial intelligence model (720).

[0193] Due to the training of the artificial intelligence model, the parameters of the classification layer changed from the initial parameters to the second-second parameters.

[0194] In addition, general training data was used during the training of the CNN, while classification training data was used during the training of the artificial intelligence model. In other words, because the data used for training are different, the 2-2 parameter set in the classification layer while training the artificial intelligence model may differ from the 2-1 parameter set in the classification layer while training the CNN.

[0195] Figure 9 is a diagram illustrating the distribution of classification layers.

[0196] The processor can transmit the updated classification layer (830) to the electronic device (710).

[0197] Specifically, the processor can separate the classification layer (830) having the second-2 parameter from the feature extraction layer and transmit it to the electronic device (710).

[0198] In this case, the electronic device (710) can receive a classification layer (830) having a second-2 parameter.

[0199] Meanwhile, the electronic device (710) is equipped with an artificial intelligence model (920) that was previously downloaded from the artificial intelligence server (100).

[0200] And the artificial intelligence model (920) installed in the electronic device (710) may include a feature extraction layer (925) having a first parameter and a classification layer (930) having a second-1 parameter. Meanwhile, the classification layer having the second-1 parameter is merely an example, and the classification layer (930) may have any other parameter.

[0201] Meanwhile, the electronic device (710) receives a classification layer (830) and can replace the classification layer (930) that was previously included in the artificial intelligence model (920) with the received classification layer (830).

[0202] And the electronic device (710) can use the feature extraction layer (925) previously included in the artificial intelligence model (920) and the received classification layer (830) by connecting them.

[0203] There is a need to update artificial intelligence models. Examples include cases where classification for new categories is required (such as updating an existing AI model that classified apples, oranges, and tomatoes to additionally classify watermelons) or cases where performance for categories frequently used by users is improved (such as updating an existing AI model that classified Black, White, and Asian people to classify Asians more effectively).

[0204] Furthermore, according to the present invention, since only the classification layer is updated and distributed, there is an advantage in that the amount of training data, training time, and the amount of distributed data can be reduced compared to a method in which both the feature extraction layer and the classification layer are updated and distributed.

[0205] FIG. 10 is a diagram illustrating a method for generating multiple classification layers and providing them to an electronic device.

[0206] The processor can generate multiple classification layers with different parameters and store them in memory (1020).

[0207] Specifically, the processor can obtain a classification layer corresponding to the first category by providing training data for classification and labeling data for classification corresponding to the first category to an artificial intelligence model and updating the classification layer, and can store the classification layer corresponding to the first category in memory. In this case, the classification layer may have a second-third parameter.

[0208] Here, the classification training data corresponding to the first category may be training data for improving the performance of an artificial intelligence model on input data of the first category.

[0209] For example, to improve the performance of an AI model on images of Koreans, the processor can update the classification layer by providing training data for classification corresponding to Koreans to the AI ​​model. Accordingly, the processor can obtain a classification layer corresponding to Koreans and store the classification layer corresponding to Koreans in memory. In this case, the classification layer may have a second-third parameter.

[0210] Additionally, the processor may obtain a classification layer corresponding to the second category by providing training data for classification and labeling data for classification corresponding to the second category to an artificial intelligence model and updating the classification layer, and may store the classification layer corresponding to the second category in the memory. In this case, the classification layer may have a second-fourth parameter different from the second-three parameter.

[0211] For example, to improve the performance of an AI model for images of Japanese people, the processor can update the classification layer by providing training data for classification corresponding to Japanese people to the AI ​​model. Accordingly, the processor can obtain a classification layer corresponding to Japanese people and store the classification layer corresponding to Japanese people in memory. In this case, the classification layer may have parameters 2-4.

[0212] Various methods for distributing classification hierarchies are explained with reference to Figs. 11 and 12.

[0213] Figure 11 is a diagram illustrating a method for replacing classification layers.

[0214] As the artificial intelligence server (100) generates and transmits an artificial intelligence model (S1105, S1110), the electronic device is in a state where the artificial intelligence model is installed.

[0215] Meanwhile, the electronic device may receive input for changing the classification layer (S1115). For example, a user who is not satisfied with the performance of the artificial intelligence model may operate their electronic device to change the classification layer. In this case, the electronic device may receive input from the user for changing the classification layer.

[0216] In addition, the electronic device can send a request to change the classification level to the artificial intelligence server (100) (S1120).

[0217] When a request to change the classification layer is received from an electronic device, the artificial intelligence server (100) can transmit a classification layer with new parameters to the electronic device (S1125).

[0218] Specifically, the artificial intelligence server (100) can transmit one or more of a plurality of classification layers having different parameters, which are stored in memory, to an electronic device. The one or more classification layers transmitted here may be those that have not previously been transmitted to an electronic device.

[0219] Meanwhile, the electronic device receives a classification layer and can replace the classification layer included in the existing artificial intelligence model with the received classification layer. Specifically, the electronic device receives the classification layer and can use the received classification layer by connecting it to the feature extraction layer included in the existing artificial intelligence model (S1130).

[0220] Meanwhile, if multiple classification layers are received, the electronic device can replace the classification layer included in the existing artificial intelligence model with the first classification layer among the received multiple classification layers.

[0221] Additionally, based on user input, the electronic device may replace the first classification layer included in the artificial intelligence model with the second classification layer among the received plurality of classification layers.

[0222] Meanwhile, the electronic device may receive input for changing the classification layer again (S1135). For example, if the user is not satisfied with the performance of the classification layer received from the artificial intelligence server, the user may operate their electronic device to change the classification layer again. In this case, the electronic device may receive input from the user for changing the classification layer again.

[0223] Then, the electronic device can send a request to change the classification hierarchy again to the artificial intelligence server (100) (S1140).

[0224] When a request to change the classification layer is received from an electronic device, the artificial intelligence server (100) can transmit a classification layer with new parameters to the electronic device (S1145).

[0225] Specifically, the artificial intelligence server (100) can transmit one or more classification layers among a plurality of classification layers having different parameters that are stored in memory to an electronic device. In addition, the artificial intelligence server (100) can transmit classification layers to an electronic device that were not previously transmitted to the electronic device.

[0226] Meanwhile, the electronic device receives a classification layer and can replace the classification layer included in the existing artificial intelligence model with the received classification layer. Specifically, the electronic device receives the classification layer and can use the received classification layer by connecting it to the feature extraction layer included in the existing artificial intelligence model (S1150). Additionally, the electronic device can delete the classification layer included in the existing artificial intelligence model from memory.

[0227] As such, the present invention has the advantage of providing an improved artificial intelligence model to a user terminal when the user is not satisfied with the current performance, by providing a classification layer having various parameters to the user terminal in response to a request to change the classification layer from the user terminal.

[0228] Figure 12 is a diagram illustrating a method for providing a classification hierarchy of categories desired by the user.

[0229] As the artificial intelligence server (100) generates and transmits an artificial intelligence model (S1205, S1210), the electronic device is in a state where the artificial intelligence model is installed.

[0230] Meanwhile, the electronic device may receive input for selecting a category (S1215). For example, a user who wants an artificial intelligence model to have high performance for a specific category may operate their electronic device to specify the category.

[0231] Meanwhile, based on the input for selecting a category, the electronic device can transmit category selection information to the artificial intelligence server (100) (S1220). Here, the category selection information may include information about what category is selected by the user.

[0232] When selection information of a category is received from an electronic device, the artificial intelligence server (100) can transmit a classification layer corresponding to the received selection information of a category to the electronic device (S1225).

[0233] Specifically, if the selection information of the received category corresponds to the first category, the artificial intelligence server (100) can transmit the classification layer corresponding to the first category to the electronic device.

[0234] For example, if the selection information of the received category represents a Korean, the artificial intelligence server (100) can transmit a classification layer corresponding to a Korean to an electronic device. For another example, if the selection information of the received category represents a Japanese, the artificial intelligence server (100) can transmit a classification layer corresponding to a Japanese to an electronic device.

[0235] Meanwhile, the electronic device receives a classification layer corresponding to the first category and can replace the classification layer included in the existing artificial intelligence model with the classification layer corresponding to the first category (S1230).

[0236] Meanwhile, electronic devices can delete classification layers included in existing artificial intelligence models from memory.

[0237] Specifically, an electronic device can obtain recognition results for input data by using an artificial intelligence model that includes a feature extraction layer and a classification layer corresponding to a first category. In this case, one or more instructions constituting the feature extraction layer and the classification layer may be stored in memory.

[0238] Meanwhile, the electronic device receives a classification layer corresponding to the second category and can replace the classification layer corresponding to the first category with the classification layer corresponding to the second category. In this case, the electronic device can delete the classification layer corresponding to the first category from memory and store the classification layer corresponding to the second category in memory.

[0239] Figure 13 is a diagram illustrating another method of providing a classification hierarchy of categories desired by the user.

[0240] The artificial intelligence server (100) has created an artificial intelligence model (S1305). However, the created artificial intelligence model has not yet been distributed to an electronic device.

[0241] Meanwhile, the electronic device may receive input for selecting a category (S1310). For example, a user who wants an artificial intelligence model to have high performance for a specific category may operate their electronic device to specify the category.

[0242] Meanwhile, based on the input for selecting a category, the electronic device can transmit category selection information to the artificial intelligence server (100) (S1315). Here, the category selection information may include information about the category selected by the user.

[0243] When selection information of a category is received from an electronic device, the artificial intelligence server (100) can transmit an artificial intelligence model including a classification layer corresponding to the received selection information of a category to the electronic device (S1320).

[0244] Specifically, the artificial intelligence server (100) can combine a classification layer corresponding to category selection information with a feature extraction layer included in the artificial intelligence model generated in S1305, and transmit the artificial intelligence model including the feature extraction layer and the classification layer to an electronic device.

[0245] Meanwhile, the electronic device receives an artificial intelligence model (S1325) and can perform recognition using the received artificial intelligence model.

[0246] As such, the present invention has the advantage of being able to provide a classification hierarchy that meets the user's needs by providing a classification hierarchy to the user terminal that exhibits high performance in a category selected by the user.

[0247] Meanwhile, although the process of updating classification layers using classification training data was previously described as being performed on an artificial intelligence server, it is not limited thereto. Specifically, if an electronic device possesses classification training data and classification labeling data, the electronic device can update classification layers by providing classification training data and classification labeling data to an artificial intelligence model to train the artificial intelligence model.

[0248] The following explains how the artificial intelligence server operates.

[0249] A method of operation of an artificial intelligence server may include the step of updating the classification layer by providing training data for classification and labeling data for classification to an artificial intelligence model that includes a feature extraction layer for extracting feature vectors and a classification layer for classifying input data using the feature vectors, thereby training the artificial intelligence model; and the step of transmitting the updated classification layer to the electronic device.

[0250] In this case, the method may further include the step of generating the artificial intelligence model comprising the feature extraction layer having the first parameter and the classification layer having the second-1 parameter by training the neural network using general-purpose training data and general-purpose labeling data.

[0251] In this case, the step of updating the classification layer may include the step of obtaining the classification layer having a 2-2 parameter different from the 2-1 parameter by providing the classification training data and classification labeling data to the artificial intelligence model and training the artificial intelligence model.

[0252] In this case, the step of obtaining the classification layer having the second-2 parameter may include replacing the classification layer having the second-1 parameter with a classification layer having an initial parameter, and training the artificial intelligence model by providing the training data for classification and the labeling data for classification to the artificial intelligence model to obtain the classification layer having the second-2 parameter.

[0253] Meanwhile, the step of generating the artificial intelligence model includes the step of adjusting the parameters of the feature extraction layer and the parameters of the classification layer so that the error between the estimated value of the neural network and the general-purpose labeling data is reduced during the process of training the neural network, and the step of obtaining the classification layer having the second-2 parameter may include the step of adjusting the parameters of the classification layer so that the error between the estimated value of the artificial intelligence model and the classification labeling data is reduced during the process of training the artificial intelligence model.

[0254] Meanwhile, the first parameter of the feature extraction layer can be maintained identically to that before the training of the artificial intelligence model, even after the artificial intelligence model has been trained.

[0255] Meanwhile, the above-mentioned trained artificial intelligence model includes the feature extraction layer having the first parameter and the classification layer having the second parameter, and the step of transmitting the updated classification layer to the electronic device may include the step of separating the classification layer having the second parameter from the feature extraction layer and transmitting it to the electronic device.

[0256] Meanwhile, the step of transmitting the updated classification layer to the electronic device may include, when a request to change the classification layer is received from the electronic device, transmitting one or more classification layers among a plurality of classification layers having different parameters to the electronic device.

[0257] Meanwhile, the step of updating the classification layer may include: a step of obtaining a classification layer corresponding to the first category by providing classification training data and classification labeling data corresponding to the first category to the artificial intelligence model to update the classification layer, and storing the classification layer corresponding to the first category in memory; and a step of obtaining a classification layer corresponding to the second category by providing classification training data and classification labeling data corresponding to the second category to the artificial intelligence model to update the classification layer, and storing the classification layer corresponding to the second category in memory.

[0258] In this case, the step of transmitting the updated classification layer to the electronic device may include receiving category selection information from the electronic device, and if the received category selection information corresponds to the first category, transmitting the classification layer corresponding to the first category to the electronic device.

[0259] The above-described invention can be implemented as computer-readable code on a medium on which a program is recorded. A computer-readable medium includes 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. Additionally, the computer may include a control unit (180) of a terminal. Accordingly, the above detailed description should not be interpreted restrictively in all respects and should be considered exemplary. The scope of the invention should 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. Explanation of the symbols

[0260] 100: AI Server

Claims

Claim 1 A communication unit that communicates with an electronic device; and a processor that updates the classification layer by providing training data for classification and labeling data for classification to an artificial intelligence model including a feature extraction layer that extracts a feature vector and a classification layer that classifies input data using said feature vector, and transmits the updated classification layer to the electronic device, wherein the processor generates the artificial intelligence model including the feature extraction layer having a first parameter and the classification layer having a second-1 parameter by training a neural network using general-purpose training data and general-purpose labeling data, and obtains the classification layer having a second-2 parameter different from the second-1 parameter by providing the training data for classification and the labeling data for classification to the artificial intelligence model to train the artificial intelligence model, and wherein the first parameter having the feature extraction layer remains the same as before the training of the artificial intelligence model even after the artificial intelligence model is trained. Claim 2 delete Claim 3 delete Claim 4 In claim 1, the processor is an artificial intelligence server that obtains a classification layer having the 2-2 parameter by replacing the classification layer having the 2-1 parameter with a classification layer having an initial parameter, and training the artificial intelligence model by providing the classification training data and the classification labeling data to the artificial intelligence model. Claim 5 In claim 1, the processor adjusts the parameters of the feature extraction layer and the parameters of the classification layer so that the error between the estimated value of the neural network and the general-purpose labeling data is reduced during the process of training the neural network, and adjusts the parameters of the classification layer so that the error between the estimated value of the artificial intelligence model and the classification labeling data is reduced during the process of training the artificial intelligence model. Claim 6 delete Claim 7 In claim 1, the trained artificial intelligence model includes the feature extraction layer having the first parameter and the classification layer having the second parameter, and the processor separates the classification layer having the second parameter from the feature extraction layer and transmits it to the electronic device. Claim 8 An artificial intelligence server according to claim 1, further comprising a memory for storing a plurality of classification layers having different parameters, wherein the processor transmits one or more of the plurality of classification layers to the electronic device when a request to change a classification layer is received from the electronic device. Claim 9 An artificial intelligence server according to claim 1, further comprising a memory for storing data, wherein the processor obtains a classification layer corresponding to the first category by providing classification training data and classification labeling data corresponding to the first category to the artificial intelligence model and updating the classification layer, and stores the classification layer corresponding to the first category in the memory, and obtains a classification layer corresponding to the second category by providing classification training data and classification labeling data corresponding to the second category to the artificial intelligence model and updating the classification layer, and stores the classification layer corresponding to the second category in the memory. Claim 10 In claim 9, the processor receives category selection information from the electronic device, and if the received category selection information corresponds to the first category, the artificial intelligence server transmits a classification hierarchy corresponding to the first category to the electronic device. Claim 11 A method of operation of an artificial intelligence server comprising: a step of updating the classification layer by providing training data for classification and labeling data for classification to an artificial intelligence model including a feature extraction layer for extracting a feature vector and a classification layer for classifying input data using said feature vector, thereby training said artificial intelligence model; and a step of transmitting said updated classification layer to an electronic device, wherein the artificial intelligence model including said feature extraction layer having a first parameter and said classification layer having a second-1 parameter is generated by training a neural network using said general-purpose training data and said general-purpose labeling data, and said classification layer having a second-2 parameter different from said second-1 parameter is obtained by providing said training data for classification and said labeling data for classification to said artificial intelligence model to train said artificial intelligence model, and said first parameter having said feature extraction layer remains the same as before the training of said artificial intelligence model even after said artificial intelligence model is trained.

Images