Electronic device for reconstructing artificial intelligence model and control method thereof

Abstract

An electronic device for reconstructing an artificial intelligence model and a control method thereof are provided. The control method comprises: inputting at least one input data into a first artificial intelligence (AI) model to obtain at least one output data; obtaining first usage information based on the obtained at least one output data; obtaining first reconstruction information for reconstructing the first AI model based on the obtained first usage information; and reconstructing the first AI model based on the obtained first reconstruction information to obtain a second AI model.

Description

Technical Field

[0001] The present disclosure relates to an artificial intelligence (AI) system and its application for simulating human brain functions such as recognition and determination by using a machine learning algorithm. More specifically, the present disclosure relates to an electronic device for reconstructing an artificial intelligence model and a control method thereof. Background Art

[0002] An artificial intelligence (AI) system is a computer system that achieves human-level intelligence and is a system that is different from previous rule-based intelligent systems in which machines can learn, determine, and become intelligent on their own. As the use of AI systems improves recognition rates and understanding of user preferences more accurately, previous rule-based intelligent systems are gradually being replaced by deep learning-based AI systems.

[0003] AI technology includes machine learning (deep learning) and component technology that uses machine learning.

[0004] Machine learning is an algorithmic technology that can classify and learn the characteristics of input data by itself. Component technology is a technology that utilizes machine learning algorithms such as deep learning, which can include various technical fields such as language understanding, visual understanding, reasoning / prediction, knowledge expression, motion control, etc.

[0005] The various fields where AI technology is applied are shown below. Language understanding is a technology that recognizes human language and characters and applies and processes them, which may include natural language processing, machine translation, dialogue systems, question and answer, speech recognition and synthesis, etc. Visual understanding is a technology that recognizes objects as if they were seen from a human's perspective, which may include object recognition, object tracking, image search, human recognition, scene understanding, spatial understanding, image improvement, etc. Reasoning / prediction is a technology that identifies information and logically infers and predicts the identified information, which may include reasoning based on knowledge and probability, optimization prediction, preference-based planning, recommendation, etc. Knowledge expression is a technology that automatically converts human experience information into knowledge data, which may include knowledge construction (data acquisition and classification), knowledge management (data utilization), etc. Motion control is a technology used to control the autonomous driving of vehicles and the movement of robots, which may include motion control (navigation, collision and driving), manipulation control (behavior control), etc.

[0006] The AI ​​technology of the prior art uses a lot of learning time and a large storage space. Therefore, the AI ​​technology of the prior art is usually performed in an external server that has a large storage space and can perform high calculations, and a method of effectively using the AI ​​model in a user terminal device such as a smart phone is being discussed.

[0007] The above information is presented as background information only to assist with an understanding of the present disclosure. No determination is made and no assertion is made as to whether any of the above content may be applicable as prior art with respect to the present disclosure. Summary of the invention

[0008] Solution

[0009] According to one aspect of the present disclosure, a control method of an electronic device includes: inputting at least one input data into a first artificial intelligence (AI) model to obtain at least one output data; obtaining first usage information based on the at least one output data obtained; obtaining first reconstruction information for reconstructing the first AI model based on the obtained first usage information; and reconstructing the first AI model based on the obtained first reconstruction information to obtain a second AI model.

[0010] The first usage information may include any one or any combination of information related to a user's usage pattern, information related to a location where the electronic device is located, age information of a user of the electronic device, context information, and environmental information.

[0011] The first reconstruction information may include any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model, and incremental information for reconstructing the first AI model, and the incremental information may include one or both of structural information of the first AI model and / or the second AI model and information about the weight difference between the first AI model and the second AI model.

[0012] The control method may also include: inputting first input data into the acquired second AI model to obtain first output data; transmitting the first input data to a server based on the reliability of the acquired first output data being less than or equal to a predetermined value; and obtaining second output data associated with the transmitted first input data from the server.

[0013] The control method may also include: obtaining second usage information based on the obtained second output data; obtaining second reconstruction information for reconstructing the second AI model based on the obtained second usage information; and reconstructing the obtained second AI model based on the obtained second reconstruction information to obtain a third AI model.

[0014] The control method may also include: based on the fact that the number of times any one of the at least one output data obtained is output is less than a predetermined value, based on the at least one output data obtained, obtaining second usage information that does not include any one of the at least one output data; based on the obtained second usage information, obtaining second reconstruction information for reconstructing the second AI model; and based on the obtained second reconstruction information, reconstructing the obtained second AI model to obtain a third AI model.

[0015] The control method may also include: displaying a user interface (UI) for adding and / or changing the acquired first usage information; modifying the acquired first usage information based on a user command for adding and / or changing the acquired first usage information being input through the displayed UI; acquiring second reconstruction information for reconstructing the second AI model based on the modified first usage information; and reconstructing the acquired second AI model based on the acquired second reconstruction information to obtain a third AI model.

[0016] Each of the first AI model and the second AI model may be used for image analysis or for speech processing and language understanding.

[0017] According to one aspect of the present disclosure, an electronic device includes: a memory storing instructions and a processor, configured to execute the stored instructions to: input at least one input data to a first artificial intelligence (AI) model to obtain at least one output data; obtain first usage information based on the at least one output data obtained; obtain first reconstruction information for reconstructing the first AI model based on the obtained first usage information; and reconstruct the first AI model based on the obtained first reconstruction information to obtain a second AI model.

[0018] The first usage information may include any one or any combination of information related to a user's usage pattern, information related to a location where the electronic device is located, age information of a user of the electronic device, context information, and environmental information.

[0019] The first reconstruction information may include any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model, and incremental information for reconstructing the first AI model, and the incremental information may include one or both of structural information of the first AI model and / or the second AI model and information about the weight difference between the first AI model and the second AI model.

[0020] The electronic device may also include a communicator, which includes a circuit, and the processor may also be configured to run the stored instructions to: input first input data to the acquired second AI model to obtain first output data; based on the reliability of the acquired first output data being less than or equal to a predetermined value, control the communicator to transmit the first input data to the server via the communicator; and obtain second output data associated with the transmitted first input data from the server via the communicator.

[0021] The processor can also be configured to run stored instructions to obtain second usage information based on the obtained second output data; obtain second reconstruction information for reconstructing the second AI model based on the obtained second usage information; and reconstruct the obtained second AI model based on the obtained second reconstruction information to obtain a third AI model.

[0022] The processor can also be configured to run the stored instructions to: based on the fact that the number of times any one of the at least one output data obtained is output is less than a predetermined value, based on the at least one output data obtained, obtain second usage information that does not include any one of the at least one output data; based on the obtained second usage information, obtain second reconstruction information for reconstructing the second AI model; and based on the obtained second reconstruction information, reconstruct the obtained second AI model to obtain a third AI model.

[0023] The electronic device may also include a display, and the processor may be further configured to run the stored instructions to: control the display to display a user interface (UI) for adding and / or changing the acquired first usage information; modify the acquired first usage information based on a user command for adding and / or changing the acquired first usage information being input through the displayed UI; obtain second reconstruction information for reconstructing the second AI model based on the modified first usage information; and reconstruct the acquired second AI model based on the acquired second reconstruction information to obtain a third AI model.

[0024] Each of the first AI model and the second AI model may be used for image analysis or for speech processing and language understanding.

[0025] According to one aspect of the present disclosure, a control method of a system includes an electronic device and a server for reconstructing an artificial intelligence (AI) model, and the control method of the system includes: inputting at least one input data into a first AI model by the electronic device to obtain at least one output data; obtaining first usage information by the electronic device based on the obtained at least one output data, and transmitting the obtained first usage information by the electronic device to the server. The control method also includes: obtaining first reconstruction information for reconstructing the first AI model by the server based on the transmitted first usage information; transmitting the obtained first reconstruction information to the electronic device by the server, and reconstructing the first AI model by the electronic device based on the transmitted first reconstruction information to obtain a second AI model.

[0026] The first usage information may include any one or any combination of information related to the user's usage pattern, information related to the location of the electronic device, age information of the user of the electronic device, context information and environmental information. The first reconstruction information may include any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model and incremental information for reconstructing the first AI model. The incremental information may include one or both of the structural information of the first AI model and / or the second AI model and the information about the weight difference between the first AI model and the second AI model.

[0027] The control method may also include: based on the fact that the number of times any one of the at least one output data obtained is output is less than a predetermined value, the electronic device obtains second usage information that does not include any one of the at least one output data based on the at least one output data obtained; the electronic device transmits the obtained second usage information to a server; the server obtains second reconstruction information for reconstructing the second AI model based on the transmitted second usage information; the server transmits the obtained second reconstruction information to the electronic device; and the electronic device reconstructs the obtained second AI model based on the transmitted second reconstruction information to obtain a third AI model.

[0028] The control method may also include: based on a user command for adding and / or changing the acquired first usage information being input through the displayed UI, the electronic device displays a user interface (UI) for adding and / or changing the acquired first usage information; modifying the acquired first usage information; transmitting the modified first usage information to a server by the electronic device; based on the transmitted first usage information, obtaining second reconstruction information for reconstructing the second AI model by the server; transmitting the acquired second reconstruction information to the electronic device by the server; and reconstructing the acquired second AI model by the electronic device based on the transmitted second reconstruction information to obtain a third AI model.

[0029] According to one aspect of the present disclosure, a system includes: an electronic device, which is configured to: input at least one input data into a first AI model to obtain at least one output data, and obtain first usage information based on the at least one output data obtained; and a server, which is configured to receive the obtained first usage information from the electronic device, obtain first reconstruction information for reconstructing the first AI model based on the received first usage information, reconstruct the first AI model based on the obtained first reconstruction information to obtain a second AI model, and transmit the obtained second AI model to the electronic device.

[0030] Each of the first AI model and the second AI model may include a neural network. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and other aspects and advantages of the embodiments of the present disclosure will become more apparent from the following description in conjunction with the accompanying drawings, in which:

[0032] Figure 1 is a diagram briefly illustrating an example of a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure;

[0033] Figure 2 is a block diagram briefly showing a configuration of an electronic device according to an embodiment of the present disclosure;

[0034] Figure 3 is a block diagram illustrating in detail the configuration of an electronic device according to an embodiment of the present disclosure;

[0035] Figure 4A is a block diagram briefly showing a configuration of an external server according to an embodiment of the present disclosure;

[0036] Figure 4B is a diagram for explaining an embodiment in which an electronic device and an external server interlock and operate with each other according to an embodiment of the present disclosure;

[0037] Figure 5A is a flowchart provided for explaining operations of an electronic device and an external server according to various embodiments of the present disclosure;

[0038] Figure 5B is a sequence diagram provided for explaining operations of an electronic device and an external server according to various embodiments of the present disclosure;

[0039] Figure 5C is a sequence diagram provided for explaining operations of an electronic device and an external server according to various embodiments of the present disclosure;

[0040] Fig. 6A is a diagram providing an example for explaining a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure;

[0041] Figure 6B is a diagram providing an example for explaining a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure;

[0042] Fig. 7A is a diagram provided for explaining an example case in which an artificial intelligence (AI) model is an image analysis model according to an embodiment of the present disclosure;

[0043] Figure 7B is a sequence diagram provided for explaining an example case in which an artificial intelligence (AI) model is an image analysis model according to an embodiment of the present disclosure;

[0044] Fig. 8Ais a diagram providing an example for explaining a method for reconstructing a language translation artificial intelligence (AI) model according to another embodiment of the present disclosure;

[0045] Figure 8B is a diagram providing an example for explaining a method for reconstructing a language translation artificial intelligence (AI) model according to another embodiment of the present disclosure;

[0046] Fig.9A are diagrams provided to explain various embodiments of the present disclosure;

[0047] Fig. 9B are diagrams provided to explain various embodiments of the present disclosure;

[0048] Fig. 9C are diagrams provided to explain various embodiments of the present disclosure;

[0049] Fig. 10A is a diagram showing an example user interface (UI) screen for inputting usage information;

[0050] Fig. 10B is a diagram showing an example user interface (UI) screen for inputting usage information;

[0051] Fig.11 is a block diagram provided for explaining the operation of an electronic device using an artificial intelligence (AI) model according to an embodiment of the present disclosure; and

[0052] Fig.12 is a flowchart provided to explain a control method of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0053] An aspect of the present disclosure is to provide a method for reconstructing an artificial intelligence (AI) model optimized for a user terminal.

[0054] Hereinafter, various embodiments of the present disclosure will be described with reference to the accompanying drawings. However, the present disclosure is not limited to the embodiments described below, but includes various modifications, equivalent scopes and / or substitutions of the embodiments of the present disclosure. Regarding the explanation of the accompanying drawings, similar figure numerals may be used for similar constituent elements.

[0055] In the specification, the terms “have”, “may have”, “include” or “may include” indicate the existence of corresponding features (eg, numerical values, functions, operations or constituent elements such as components), but do not exclude the existence of other features.

[0056] In the specification, the term "A or B", "at least one of A or / and B", or "one or more of A or / and B" may include all possible combinations of the items listed together. For example, the term "A or B" or "at least one of A or / and B" may mean (1) at least one A, (2) at least one B, or (3) at least one A and at least one B.

[0057] In the specification, the terms “first, second, etc.” are used to describe various elements without considering their order and / or importance and to distinguish one element from other elements, but are not limited to the corresponding elements.

[0058] If it is described that one element (e.g., a first element) is "operably or communicatively coupled" or "connected to" another element (e.g., a second element), the element may be connected to the other element directly or through another element (e.g., a third element). Conversely, it will be understood that when an element (e.g., a first element) is "directly coupled / coupled to" or "directly connected to" another element (e.g., a second element), there is no intervening element (e.g., a third element) between the element and the other element.

[0059] The expression "configured to (or set to)" used in one or more embodiments of the present disclosure may be replaced with "suitable for", "capable of", "designed to", "suitable for", "made to", or "capable of" depending on the context. The wording "configured to (set to)" does not necessarily mean or refer to "specially designed to" at the hardware level. On the contrary, in some cases, the term "device configured to (execute)" may mean that the device is able to "execute" certain operations with other devices or components. For example, the phrase "a processor configured (or set to) execute A, B, and C" may mean or refer to a dedicated processor (e.g., an embedded processor) for executing the corresponding operations or a general-purpose processor (e.g., a CPU or application processor) that can perform the corresponding operations by executing one or more software programs stored in a storage device.

[0060] The electronic device according to various embodiments may include, for example, any one or any combination of a smart phone, a tablet computer, a mobile phone, a video phone, an e-book reader, a desktop computer, a laptop computer, a netbook computer, a workstation, a server, a PDA, a portable multimedia player (PMP), an MP3 player, a medical device, a camera, and a wearable device. In some embodiments, the electronic device may include, for example, any one or any combination of a television, a digital video disc (DVD) player, an audio, a refrigerator, an air conditioner, a cleaner, an oven, a microwave oven, a washing machine, an air cleaner, a set-top box, a home automation control panel, a security control panel, a media box (e.g., Samsung HomeSyncTM, Apple TVTM, or Google TVTM), a game console (e.g., XboxTM, PlayStationTM), an electronic dictionary, an electronic key, a portable camera, and an electronic frame (e-frame).

[0061] In the present disclosure, the term "user" may indicate a person using an electronic device or a device (eg, an artificial intelligence electronic device) using an electronic device.

[0062] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings.

[0063] Figure 1 is a diagram briefly illustrating an example of a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure.

[0064] like Figure 1 As shown in the left figure of , the model initially included in the electronic device 100 may generally be a general artificial intelligence (AI) model 10. That is, the AI ​​model that has been included in the electronic device 100 since the electronic device 100 was started may generally be a general AI model 10, which is used to output stable result values ​​for various input data. Therefore, the general AI model 10 includes a node or layer for outputting the result value of the input data that is not used by the user, thereby using unnecessary resources, or using unnecessary resources to determine the characteristics of the unused input data.

[0065] Therefore, the electronic device 100 can reconstruct the general AI model 10 based on the user's usage information, thereby preventing unnecessary use of resources. Figure 1 As shown in the right figure of , the electronic device 100 can reconstruct the general AI model 10 by deleting at least one layer from the multiple layers included in the general AI model 10, removing non-essential parts of at least one layer, or adding new nodes to at least one layer.

[0066] In one embodiment, in the case where the general AI model 10 is an image analysis model and the electronic device 100 only identifies a specific object (e.g., eggs, milk, carrots, etc.), the electronic device 100 may focus on information for identifying the specific object and reconstruct the general AI model 10. In another embodiment, in the case where the general AI model 10 is a language translation model, the electronic device 100 may reconstruct the general AI model 10 based only on information for translating languages ​​in a specific context (e.g., a conversation about a travel destination, a business conversation, etc.). In another embodiment, in the case where the general AI model 10 is a speech identification model, the electronic device 100 may reconstruct the general AI model 10 based only on information for identifying specific words (e.g., words frequently used by the user) or specific users (e.g., users who frequently use the electronic device 100). In another embodiment, in the case where the general AI model 10 is an AI model for removing noise, the electronic device 100 may reconstruct the general AI model 10 based on specific environmental information (e.g., structural information, size information, etc. of the room where the electronic device 100 is located).

[0067] Through this method, the electronic device 100 can reconstruct the general AI model 10 to obtain the personalized AI model 20. The personalized AI model 20 may be lighter than the general AI model 10, or the processing speed may be faster.

[0068] Figure 2 is a block diagram briefly showing a configuration of an electronic device according to an embodiment of the present disclosure.

[0069] like Figure 2 As shown, the electronic device 100 may include a memory 110 and a processor 120 .

[0070] The memory 110 may store one or more instructions or data about any one or any combination of the other elements of the electronic device 100. The memory 110 may be implemented as a non-volatile memory, a volatile memory, a flash memory, a hard disk drive (HHD), or a solid state drive (SSD). The memory 110 may be accessed by the processor 120, which may perform reading / recording / revision / deletion / update of data. According to an embodiment of the present disclosure, the term of the memory may include the memory 110, a read-only memory (ROM) and a random access memory (RAM) within the processor 120, and a memory card (e.g., a micro secure digital (SD) card or a memory stick) attached to the electronic device 100.

[0071] The memory 110 may store an AI model, output data obtained based on input data input to the AI ​​model, usage information obtained based on the output data, and reconstruction information for reconstructing the AI ​​model obtained based on the usage information. In this case, the AI ​​model may be constructed by considering the application field of the identification model, the learning object, the computer performance of the device, etc. In addition, the AI ​​model may be, for example, a model based on a neural network. The AI ​​model may be designed to simulate the structure of the human brain on a computer. The AI ​​model may include a plurality of network nodes with weights to simulate neurons of a human neural network. A plurality of network nodes may form connection relationships, respectively, thereby simulating synaptic activity in which neurons exchange signals through synapses. The AI ​​model may include, for example, a neural network model or a deep learning model developed from a neural network model. In a deep learning model, a plurality of network nodes may be located at different depths (or layers) and exchange data according to a convolutional connection relationship. For example, models such as a deep neural network (DNN), a recursive neural network (RNN), and a bidirectional recursive deep neural network (BRDNN) may be used as an artificial intelligence (AI) model, but examples are not limited thereto.

[0072] The processor 120 may be electrically connected to the memory 110 and control overall operations and functions of the electronic device 100 .

[0073] For example, the processor 120 may obtain at least one output data in association with at least one input data input into a first artificial intelligence (AI) model, obtain usage information based on the at least one output data obtained, obtain reconstruction information for reconstructing the AI ​​model based on the obtained usage information, and reconstruct the first AI model based on the reconstruction information to obtain a second artificial intelligence (AI) model.

[0074] In this case, the usage information may include any one or any combination of information related to the user's usage pattern, information related to the location where the electronic device 100 is located, age information of the user of the electronic device 100, context information, information stored in the electronic device 100, and environmental information. In this case, the information related to the user's usage pattern may include information related to the usage behavior that occurs when the user uses the electronic device 100 under predetermined conditions (e.g., time, place, etc.). In addition, the information related to the location where the electronic device 100 is located may include global positioning system (GPS) information of the electronic device 100, context information related to the GPS information (e.g., context information such as a pharmacy, a shopping mall, or a conference room). The context information may be information about various contexts. For example, the context information may be various information related to a specific place or event, such as information related to travel, information related to a pharmacy, etc. In this case, the context information may include different information according to the type of AI model. For example, in the case where the AI ​​model is an image analysis model, the context information related to travel may be information related to tourist attractions. As another example, in the case where the AI ​​model is a language translation model, the context information related to travel may be information about words frequently used in travel destinations. The information stored in the electronic device 100 may include, for example, an address book, a music list, emails, short message service (SMS) messages, etc. The environmental information may include, for example, any one or any combination of weather information, temperature information, structural information, and size information of the location where the electronic device 100 is located.

[0075] For example, the usage information may represent various information according to the type and nature of the AI ​​model. For example, in the case where the AI ​​model is an AI model for image analysis, the usage information may be information related to an object included in an image acquired through image analysis. As another example, in the case where the AI ​​model is a model for language translation, the usage information may include any one or any combination of location information (e.g., GPS information) of the electronic device 100, context information related to the location (e.g., context information such as a pharmacy, a shopping mall, or a conference room), language information related to the location (e.g., information related to the language used at the corresponding location), and schedule information stored in the electronic device 100. As another example, in the case where the AI ​​model is an AI model included in a dialogue system, the usage information may be category information for classifying text acquired through voice identification, or category information for classifying user intent (or task or command) acquired through natural language. As another example, in the case where the AI ​​model is a voice identification model, the usage information may be user identification information for identifying a user who has expressed an input voice. As another example, in the case where the AI ​​model is an AI model for processing noise, the usage information may be information related to the structure and size of the place where the electronic device 100 is located.

[0076] The reconstruction information may include any one or any combination of information related to the structure of the second AI model, information related to the weight of the second AI model, and incremental information for reconstructing the first AI model. The information related to the structure of the AI ​​model may represent information related to the structure of the layers included in the AI ​​model or information related to the connection relationship between the layers. The incremental information may include one or both of the structural information of the first AI model and the reconstructed second AI model and information related to the difference between the weights of the first AI model and the reconstructed second AI model.

[0077] In the case of acquiring the second AI model, the processor 120 may delete the first AI model from the memory 110. For example, in the case of changing the first AI model to the second AI model and reconstructing, the processor 120 may delete the data related to the first AI model. In the case of acquiring a new second AI model from the first AI model, the processor 120 may delete the previous first AI model, so that the memory 110 can be used efficiently.

[0078] As described above, the processor 120 may reconstruct the first AI model to obtain the second AI model. That is, the first AI model is a general AI model and may obtain output data about various input data. However, it has a large amount of calculation and a large size of the AI ​​model. In contrast, the reconstructed second AI model is a personalized AI model, which may output output data more accurately in association with the input data related to the usage information, and may be a lightweight AI model.

[0079] For example, in the case where the reliability of the first output data is less than or equal to a predetermined value, the processor 120 may obtain first output data about the first input data input to the reconstructed second AI model, and then transmit the first input data to the external server 200, and obtain second output data about the first input data from the external server 200. That is, in the case where the recognition probability value of the first output data obtained by inputting the first input data to the reconstructed second artificial intelligence model is less than or equal to a threshold value or the first output data indicating that the first input data is not recognized is obtained, the processor 120 may obtain output data (second output data) about the first input data from the external server 200. Through the above processing, in the case where input data unrelated to the usage information is input and incorrect output data (first output data) is obtained, the processor 120 can obtain the second output data about the first input data through the general AI model (e.g., the first AI model) of the external server 200.

[0080] In addition, the processor 120 may acquire second usage information based on the second output data acquired from the external server 200, acquire second reconstruction information for reconstructing the second UI model based on the acquired second usage information, and reconstruct the second UI model based on the second reconstruction information to acquire a third AI model. That is, in the case where the first input data that does not correspond to the previously acquired first usage information is input, the processor 120 may acquire the second usage information including the first input data and reconstruct the AI ​​model.

[0081] In addition, when the number of outputs of any one of the at least one output data under a predetermined condition (e.g., time or location) is less than a threshold, the processor 120 may obtain third usage information other than any one of the at least one output data, obtain third reconstruction information for reconstructing the second AI model based on the obtained third usage information, and reconstruct the second UI model based on the third reconstruction information to obtain a fourth AI model. That is, in contrast to when the third AI model is obtained, the processor 120 may obtain third usage information other than output data about unused input data and reconstruct the AI ​​model. Through the above-mentioned various reconstruction processes, the processor 120 can obtain a personalized and mitigated AI model.

[0082] As described above, the AI ​​model may be an AI model according to various purposes. For example, the AI ​​model may include various types of AI models, such as an AI model for image analysis, an AI model for language translation, an AI model for speech recognition, an AI model for natural language processing, an AI model for noise removal, etc.

[0083] Figure 3 is a block diagram illustrating in detail the configuration of an electronic device according to an embodiment of the present disclosure.

[0084] like Figure 3 As shown, in addition to the memory 110 and the processor 120, the electronic device 100 may further include a communicator including a circuit 130, an input interface 140, a display 150, and an audio output interface 160. However, examples are not limited to the above elements, and some elements may be added or omitted as needed.

[0085] The communicator 130 is a configuration for performing communication with another electronic device. The communicator 130 can be communicatively connected to another electronic device via a third device (e.g., a repeater, a hub, an access point, a server, a gateway, etc.). Wireless communication may include, for example, cellular communication using any one or any combination of 5G networks, LTE, LTE evolution (LTE-A), code division multiple access (CDMA), wideband CDMA (WCDMA), universal mobile telecommunications system (UMTS), wireless broadband (WiBro), global mobile communication system (GSM), etc. According to one embodiment, wireless communication may include, for example, any one or any combination of wireless fidelity technology (WiFi), Bluetooth, Bluetooth low energy (BLE), Zigbee, near field communication (NFC), electromagnetic security transmission, radio frequency (RF) and human body area network (BAN). Wired communication may include, for example, any one or any combination of universal serial bus (USB), high-definition multimedia interface (HDMI), recommended standard 232 (RS-232), power line communication and plain old telephone service (POTS). The network through which wired or wireless communications are performed may include a telecommunication network, for example, any one or any combination of a computer network (eg, a LAN or a WAN), the Internet, and a telephone network.

[0086] The communicator 130 may acquire output data regarding input data from the external server 200. In addition, in the case of reconstruction of the AI ​​model operated in the external server 200, the electronic device 100 may acquire reconstruction information or the reconstructed AI model only through the communicator 130.

[0087] The input interface 140 may be configured to obtain input of user commands. The input interface 140 may include a camera 141, a microphone 142, a touch panel 143, and the like. The camera 141 may be configured to obtain image data around the electronic device 100. The camera 141 may capture still images and moving images. For example, the camera 141 may include one or more image sensors (e.g., a front sensor or a rear sensor), a lens, an image signal processor (ISP), or a flash (e.g., an LED or a xenon lamp). The image data obtained by the camera 141 may be input as input data of the AI ​​model. The microphone 142 may be configured to obtain sounds around the electronic device 100. The microphone 142 may obtain input of an external acoustic signal and obtain electrical voice information. The microphone 142 may utilize various noise removal algorithms to remove noise that occurs during the process of obtaining input of an external acoustic signal. The audio data obtained by the microphone 142 may be input as input data of the AI ​​model. The touch panel 143 may be configured to obtain various user inputs. The touch panel 143 may obtain input of data through user manipulation. The touch panel 143 may be configured in combination with a display to be described later. The user command input through the touch panel 143 may be a command for inputting usage information. In addition to the above-mentioned camera 141, microphone 142 and touch panel 143, the input interface 140 may be configured as various elements for inputting various data.

[0088] The display 150 may be configured to output various images. The display 150 for providing various images may be implemented as a display panel in various forms. For example, the display panel may be implemented with various display technologies, such as a liquid crystal display (LCD), an organic light emitting diode (OLED), an active matrix organic light emitting diode (AM-OLED), a liquid crystal on silicon (LcoS), or a digital light processing (DLP). The display 150 may be implemented in the form of a flexible display, which may be combined with any one or any combination of the front surface area, the side surface area, and the rear surface area of ​​the electronic device 100.

[0089] The display 150 may display a user interface (UI) for adding or changing usage information. In this case, the usage information that can be added through the UI is usage information that cannot be obtained from the output data, which may be, for example, information about the usage pattern of the electronic device 100, information about the location of the electronic device 100, age information of the user of the electronic device, any one or any combination of context information and environmental information. When a user command for adding new usage information is input through the UI displayed on the display 150, the processor 120 may obtain fourth reconstruction information for reconstructing the second UI model based on the new usage information, and reconstruct the second UI model based on the fourth reconstruction information to obtain a fifth AI model. As another example, the electronic device 100 may delete usage information that is not used as output data according to the user input obtained through the UI.

[0090] The audio output interface 160 may be configured to not only output various audio data to which various processing operations such as decoding, amplification, and noise filtering are performed by an element (e.g., a CPU or dedicated hardware) for performing audio processing, but also output various alarm sounds or voice messages. The audio processing component is an element for processing audio data. The audio processing component may be associated with the audio data to perform various processing, such as decoding, amplification, noise filtering, etc. The audio data processed by the audio processing component may be output to the audio output interface 160. The audio output interface may be implemented as a speaker; this is one of the various embodiments of the present disclosure. The audio output interface may be implemented as an output component that can output audio data. In the case where the output data of the AI ​​model is output as audio, the audio output interface 160 may output audio corresponding to the output data.

[0091] The processor 120 may be configured to control the overall operation of the electronic device 100, as described above. The processor 120 may include a RAM 121, a ROM 122, a graphics processing component 123, a main central processing unit (CPU) 124, a first interface 125-1 to an nth interface 125-n, and a bus 126. The RAM 121, the ROM 122, the graphics processing component 123, the main central processing unit (CPU) 124, the first interface 125-1 to the nth interface 125-n may be connected to each other via the bus 126.

[0092] The ROM 122 may store an instruction set for system startup. When a power-on command is input and power is supplied, the main CPU 124 may copy the O / S stored in the memory to the RAM 121 according to the instructions stored in the ROM 122, and start the system by running the O / S. When the startup is completed, the main CPU 124 may copy various application programs stored in the memory to the RAM 121, run the application programs copied to the RAM 121, and perform various operations.

[0093] For example, the main CPU 124 may access the memory 110 and perform booting using the O / S stored in the memory 110. The main CPU 124 may perform various operations using various programs, contents, data, etc. stored in the memory 110.

[0094] The first to nth interfaces 125-1 to 125-n may be connected to the above-mentioned respective elements. One of the interfaces may be a network interface connected to an external device via a network.

[0095] Figure 4A is a block diagram briefly illustrating a configuration of an external server according to an embodiment of the present disclosure. Figure 4B is a diagram provided to explain an embodiment in which an electronic device and an external server interlock and operate with each other according to an embodiment of the present disclosure.

[0096] like Figure 4A As shown, the external server 200 may include a memory 210, a communicator including a circuit 220, and a processor 230. The structures, functions, and operations of the memory 210 and the communicator 220 are similar to those of the memory 110 and the communicator 130 of the electronic device 100 described above.

[0097] That is, the memory 210 may store one or more instructions or data related to at least one other element of the server 200. For example, the memory 210 may be implemented as a non-volatile memory, a volatile memory, a flash memory, a hard disk drive (HDD), a solid state drive (SSDI), etc. The memory 210 may be accessed by the processor 230, and the processor 230 may perform operations such as reading, writing, modifying, deleting, and updating data.

[0098] The memory 210 may store information about the first AI model as a general AI model. In addition, in the case where the electronic device 100 uses the first AI model stored in the server 200 to obtain output data, the memory 210 may store output data obtained based on input data input to the first AI model, usage information obtained based on the output data, and reconstruction information for the reconstructed AI model obtained based on the usage information. In addition, the memory 210 may also include usage information, reconstruction information, etc. about the second AI model reconstructed by the fourth AI model. In this case, in the case of acquiring the second AI model to the fourth AI model from the electronic device 100, the memory 210 may acquire the second AI model to the fourth AI model from the electronic device 100 and store them. However, in the case of acquiring the second AI model to the fourth AI model from the external server 200, the memory 210 may store the acquired second AI model to the fourth AI model.

[0099] The communicator 220 is a configuration for communicating with another electronic device. The communicator 220 may be communicatively connected to another electronic device via a third device (e.g., a repeater, a hub, an access point, a server, a gateway, etc.). The communicator 220 may communicate with an external device through the various wired or wireless communication schemes described above. For example, the communicator 220 may obtain first input data from the electronic device 100. However, the example is not limited thereto, and as described above, the communicator 220 may also obtain the second AI model to the fourth AI model, usage information about the second AI model to the fourth AI model, reconstruction information, etc.

[0100] The processor 230 may be electrically connected to the memory 210 , and control overall operations and functions of the external server 200 .

[0101] For example, in the case of obtaining a control signal requesting first input data and output data about the first input data from the electronic device 100, the processor 230 may obtain second output data output by inputting the first input data into the first AI model. The processor 230 may transmit the obtained second output data to the electronic device 100 via the communicator 220. That is, in the case where the electronic device 100 only stores the second AI model and cannot obtain accurate output data associated with the first input data, the processor 230 may obtain second output data that is accurate output data associated with the first input data by using the first AI model instead of the electronic device 100, and transmit the obtained second output data to the electronic device 100.

[0102] As described above, the external server 200 may perform all or part of the operations performed by the electronic device 100 instead of the electronic device 100. Figure 4B As shown, the electronic device 100 and the external server 200 may interlock and operate with each other.

[0103] In one embodiment, the electronic device 100 may acquire at least one output data associated with at least one input data input to the first AI model, and transmit first usage information acquired based on the acquired output data to the external server 200. The external server 200 may acquire first reconstruction information for reconstructing the first AI model based on the first usage information acquired from the electronic device 100, and transmit the acquired first reconstruction information to the electronic device 100. The electronic device 100 may reconstruct the first AI model based on the acquired first reconstruction information to acquire a second AI model.

[0104] In the case where the first input data that cannot be identified by the reconstructed second AI model is input to the electronic device 100, the external server 200 may obtain the first input data and obtain the second reconstruction information. For example, when the first input data is input to the second AI model of the electronic device 100, the electronic device 100 may identify the reliability of the first output data associated with the first input data. If the reliability of the first output data is less than or equal to a predetermined value (or, if the first output data indicating the unrecognized first input data is output), the electronic device 100 may transmit the first input data to the external server 200. The external server 200 may input the acquired first input data to the first AI model of the external server 200, and obtain the second output data, obtain the second usage information based on the second output data, and obtain the second reconstruction information based on the second usage information, and transmit the acquired second reconstruction information to the electronic device 100. The electronic device 100 may reconstruct the second AI model based on the acquired second reconstruction information to obtain a third AI model.

[0105] In the case where the first input data that cannot be identified by the reconstructed second AI model is input to the electronic device 100, the external server 200 may obtain the first input data and transmit the second output data associated with the obtained first input data to the electronic device 100. That is, compared with the electronic device 100 reconstructing the second AI model into an AI model capable of identifying the first input data and obtaining the second output data associated with the first input data, the external server 200 transmits the second output data associated with the first input data to the electronic device 100 faster and more accurately. Therefore, the external server 200 may transmit the second output data to the electronic device 100. The electronic device 100 may output the obtained second output data.

[0106] Figure 5A is a flowchart provided to explain operations of an electronic device and an external server according to various embodiments of the present disclosure. Figure 5B is a sequence diagram provided to explain operations of an electronic device and an external server according to various embodiments of the present disclosure. Figure 5C is a sequence diagram provided to explain operations of an electronic device and an external server according to various embodiments of the present disclosure.

[0107] like Figure 5A As shown, the electronic device 100 can obtain output data output by inputting input data into the AI ​​model, usage information obtained based on the output data, and reconstruction information obtained based on the usage information, and reconstruct the AI ​​model based on the obtained reconstruction information.

[0108] That is, in operation S511, the electronic device 100 may obtain output data associated with input data. For example, the electronic device 100 may input the input data into the first AI model to obtain output data.

[0109] In operation S512, the electronic device 100 may obtain usage information from the output data. As described above, the usage information may include any one or any combination of information about the user's usage pattern, information about the location of the electronic device 100, age information of the user of the electronic device, context information, and environmental information.

[0110] In operation S513, the electronic device may identify whether the usage information satisfies a predetermined condition or a current condition. The predetermined condition may indicate a condition for obtaining the usage information for obtaining the reconstruction information from the plurality of pieces of usage information obtained by the electronic device 100. For example, the predetermined condition may include various conditions such as a condition that the output of specific output data is less than a threshold value within a predetermined time period, a condition that the output of the output data is greater than or equal to a threshold value when the electronic device 100 exists at a predetermined position, a condition that the specific output data is set by a user, and the like.

[0111] In the case where the usage information does not satisfy the predetermined condition (S513-N), the electronic device 100 may return to operation S511 and obtain output data about the input data until the predetermined condition is satisfied. In the case where the usage information satisfies the predetermined condition (S513-Y), in operation S514, the electronic device 100 may obtain reconstruction information from the usage information that satisfies the predetermined condition. For example, when acquiring the usage information, the output data corresponding to 20 categories out of 1000 categories that can be output by the first AI model as a general AI model comes from the usage information, and the electronic device 100 may obtain reconstruction information for reconstructing the first AI model into a second AI model capable of outputting 20 categories. As described above, the reconstruction information is information for reconstructing the AI ​​model, which may include any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model, and incremental information for reconstructing the first AI model.

[0112] In operation S515 , the electronic device 100 may reconstruct the AI ​​model based on the acquired reconstruction information.

[0113] like Figure 5B and Figure 5C As shown, Figure 5A Some operations of the electronic device 100 may also be performed in the external server 200.

[0114] That is, if Figure 5B As shown, the external server 200 may acquire reconstruction information, reconstruct AI from the acquired reconstruction information, and transmit the reconstructed AI model to the electronic device 100 .

[0115] For example, in operation S521, the electronic device 100 may obtain output data associated with the input data. In operation S522, the electronic device 100 may obtain usage information from the output data. When the usage information satisfies a predetermined condition or a current condition, in operation S523, the electronic device 100 may transmit the obtained usage information satisfying the predetermined condition or the current condition to an external server. When the predetermined condition is not satisfied, the electronic device 100 may send the usage information to the external server as in Figure 5A As in operation S513-N, output data can be obtained until a predetermined condition is obtained. In operation S523, the external server 200 may obtain an AI model stored in the electronic device 100. That is, the external server 200 may obtain an AI model to be reconstructed from the electronic device 100. However, the example is not limited thereto, and the external server 200 may only obtain information about the AI ​​model to be reconstructed from the electronic device 100.

[0116] In operation S524, the external server 200 may acquire reconstruction information from the usage information acquired from the electronic device 100. In operation S525, the external server 200 may reconstruct the AI ​​model based on the acquired reconstruction information. In operation S526, the external server 200 may transmit the reconstructed AI model to the electronic device 100. In this case, the electronic device 100 may delete the previous AI model and store the AI ​​model acquired from the external server 200.

[0117] In another embodiment, if Figure 5C As shown, in operation S531, the electronic device 100 may obtain output data associated with the input data. In operation S532, the electronic device 100 may obtain usage information from the output data. When the usage information satisfies a predetermined condition or a preset condition, in operation S533, the electronic device 100 may transmit the obtained usage information that satisfies the predetermined condition or the preset condition to an external server. In operation S534, the external server 200 may obtain reconstruction information from the usage information obtained from the electronic device 100. In operation S535, the external server 200 may transmit the obtained reconstruction information to the electronic device 100. In operation S536, the electronic device 100 may reconstruct the AI ​​model based on the obtained reconstruction information.

[0118] That is, with Figure 5B Different, in Figure 5C In the embodiment, the electronic device 100 can obtain reconstruction information from the external server 200 and directly reconstruct the AI ​​model.

[0119] remove Figure 5B and Figure 5CIn addition to the embodiments, the external server 200 may perform any one or any combination of an operation of inputting input data into the AI ​​model and obtaining output data, an operation of obtaining usage information (or context information, environmental information, etc.) based on the output data, an operation of obtaining reconstruction information based on the usage information, and an operation of reconstructing the AI ​​based on the reconstruction information. For example, when the AI ​​model is to be used without storing it in the electronic device 100, the electronic device 100 may initially transmit the input data to the external server 200 and obtain the output data from the external server 200. The server 200 may obtain usage information based on the output data corresponding to the input data obtained from the electronic device 100, obtain reconstruction information based on the usage information, and reconstruct the AI ​​model based on the reconstruction information. The external server 200 may transmit the reconstructed AI model to the electronic device 100, and the electronic device 100 may use the reconstructed AI model obtained from the external server 200.

[0120] Fig. 6A is a diagram providing an example for explaining a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure. Figure 6B is a diagram providing an example for explaining a method for reconstructing an artificial intelligence (AI) model according to an embodiment of the present disclosure.

[0121] exist Fig. 6A The AI ​​model shown in 610 may be an AI model for image analysis. The electronic device 100 may acquire input data 610. In this case, the input data may be still image data including at least one object. However, the example is not limited thereto, and the input data may be moving image data.

[0122] The electronic device 100 may input a plurality of input data 610 to the first AI model 615 under predetermined conditions (e.g., a predetermined time or location) and obtain a plurality of output data. The first AI model 615 is an artificial intelligence (AI) model capable of outputting output data of 1000 categories (i.e., an artificial intelligence model capable of recognizing 1000 objects), which may include N layers (e.g., core layers, convolutional layers, pooling layers, local connection layers, recurrent layers, embedding layers, merging layers, advanced activation layers, normalization layers, noise layers, etc.) and a first weight value corresponding to each of the N layers, and may include 1000 classifier nodes for classifying the 1000 categories. In addition, the first AI model 615 is output data, which may output information about objects included in the image data.

[0123] For example, the plurality of output data acquired by the first AI model 615 under predetermined conditions may be 20 categories of output data 620. The 20 categories of output data may include information about 20 kinds of objects. That is, the electronic device 100 may identify 20 objects through the first AI model 615 under predetermined conditions. In this case, the electronic device 100 may store the output data acquired under predetermined conditions in the form of a histogram.

[0124] The electronic device 100 may obtain usage information 625 about 20 categories based on the output data 620 of the 20 categories. For example, the electronic device 100 may obtain usage information about categories (e.g., types of objects) recognized at least a predetermined number of times (e.g., twice) using a histogram stored in the electronic device 100.

[0125] The electronic device 100 may obtain reconstruction information 630 based on the acquired usage information 625. The reconstruction information 630 may be any one or any combination of structural information (e.g., number of layers) of a second AI model capable of outputting 20 categories of output data, a weight value or an incremental value of a previous first AI model 615, and an AI model capable of outputting 20 categories of output data. For example, the reconstruction information 630 may include N-2 layers, and include weight values ​​corresponding to the N-2 layers, respectively. The reconstruction information 630 may be information about an AI model capable of including 20 classifier nodes. The reconstruction information 630 may be acquired by the electronic device 100. However, the example is not limited thereto, and it may be acquired by the external server 200. That is, the electronic device 100 may transmit 20 categories of usage information to the external server 200, and obtain reconstruction information obtained based on the 20 categories of usage information from the external server 200.

[0126] The electronic device 100 may reconstruct the first AI model 615 into the second AI model 635 based on the reconstruction information. The second AI model 635 may include fewer layers than the first AI model 615 and include fewer classifier nodes, and thus may be lighter than the first AI model 615. For example, the second AI model 635 may include N-2 layers, include second weight values ​​corresponding to the N-2 layers, respectively, and include 20 classifier nodes.

[0127] In this case, the electronic device 100 may delete the file storing information about the first AI model 615 stored in the memory 110, and store a file storing information about the second AI model 635 (that is, reconstruction information) in the memory 110. Alternatively, the electronic device 100 may update the previously executed first AI model 615 to the second AI model 635 based on the reconstruction information.

[0128] Figure 6BThe AI ​​model shown in 640 may be an AI model for language translation among AI models for speech processing and language understanding. The electronic device 100 may acquire input data 640. The input data may be voice data of a user.

[0129] The electronic device 100 may input a plurality of input data 640 to the first AI model 645 under a predetermined condition (e.g., a predetermined time or location) and obtain a plurality of output data. The first AI model 645 may be an AI model for language translation related to a hospital, which may include an embedding layer including a list of N words related to the hospital. In addition, the first AI model 645 is output data, which may output translation information about the user's voice.

[0130] For example, some of the plurality of output data acquired by the first AI model 645 under a predetermined condition may acquire output data 650 about a shopping mall. For example, the electronic device 100 may acquire text related to a shopping mall (eg, the name, location, business hours of the shopping mall) as output data.

[0131] Alternatively, the electronic device 100 may acquire context information 655 related to a shopping mall. For example, the electronic device 100 may acquire, as the context information 655, any one or any combination of location information (e.g., GPS information) of the electronic device 100, context information related to the location (e.g., whether the electronic device is located in a shopping mall, etc.), language information related to the location (e.g., information about the language used at the corresponding location), schedule information stored in the electronic device 100, and information set by the user through the UI.

[0132] The electronic device 100 may obtain usage information related to shopping 660 based on some of the output data 650 output by the first AI model 645 and the context information 655. In this case, the electronic device 100 may obtain usage information related to shopping 660 based on one of the output data 650 output by the first AI model 645 or the context information 655.

[0133] The electronic device 100 may acquire reconstruction information 665 based on the acquired usage information 660. The reconstruction information 665 may be any one or any combination of structural information (e.g., number of layers) of a second AI model capable of performing language translation related to shopping, a weight value, or an incremental value of a previous first AI model 645, and an AI model capable of outputting language translation related to shopping. For example, the reconstruction information 665 may be information about a second AI model 670 including an embedding layer including M lists of words related to shopping. In this case, the second AI model 670 may include a different number of layers or a different weight value than the first AI model 645.

[0134] The reconstruction information 665 may be acquired by the electronic device 100. However, the example is not limited thereto, and it may also be acquired by the external server 200. That is, the electronic device 100 may transmit the usage information related to shopping to the external server 200, and acquire the reconstruction information acquired based on the usage information related to shopping from the external server 200.

[0135] The electronic device 100 may reconstruct the first AI model 645 into the second AI model 670 based on the reconstruction information 665. The second AI model 670 may include an embedding layer including a different word list from the first AI model 645 and including a different number of layers from the first AI model 645. That is, the electronic device 100 may perform language translation using an AI model optimized according to the user's usage information.

[0136] In this case, the electronic device 100 may delete the file storing information about the first AI model 645 stored in the memory 110, and store the file storing information about the second AI model 670 (that is, reconstruction information) in the memory 110. Alternatively, the electronic device 100 may update the previously executed first AI model 645 to the second AI model 670 based on the reconstruction information. Fig. 7A , Figure 7B , Fig. 8A , Figure 8B , Fig.9A , Fig. 9B and Fig. 9C is a diagram provided for explaining an example method for reconstructing an artificial intelligence (AI) model according to various embodiments of the present disclosure.

[0137] Fig. 7A is a diagram provided for explaining an example case in which an artificial intelligence (AI) model is an image analysis model according to an embodiment of the present disclosure. Figure 7B is a sequence diagram provided for explaining an example case in which an artificial intelligence (AI) model is an image analysis model according to an embodiment of the present disclosure.

[0138] like Fig. 7A As shown, the electronic device 100 may be a refrigerator, and the AI ​​model may be an image analysis model for identifying the type of products inside the refrigerator.

[0139] The electronic device 100 may use the first AI model to acquire data about various types of products stored in the electronic device 100. For example, the electronic device 100 may input image data generated by capturing various types of food to the first AI model and acquire information about the product as output data.

[0140] When the acquired output data satisfies a predetermined condition, the electronic device 100 may acquire usage information based on the output data that satisfies the predetermined condition, and reconstruct the first AI model based on the acquired usage information to acquire a second AI model. The predetermined condition may be, for example, capturing a product more than a predetermined number of times within a predetermined time period. Fig. 7A As shown in the left figure of , the electronic device 100 can obtain information about the first product 701 to the third product 703 existing inside the electronic device 100 through the first AI model. The electronic device 100 can obtain usage information based on the information about the first product 701 to the third product 703, obtain reconstruction information for identifying the first product 701 to the third product 703 based on the obtained usage information, and obtain a second AI model reconstructed based on the obtained reconstruction information.

[0141] New products may be added to the electronic device 100 according to the user's use, or products may be consumed and disappear. That is, a fourth product 704 that satisfies a predetermined condition may be added according to the user's use of the electronic device 100, and there may be products that previously satisfied the predetermined condition but do not currently satisfy the condition. That is, Fig. 7A As shown in the upper right figure of , the electronic device 100 can reconstruct the second AI model to identify the fourth product 704, or as Fig. 7A As shown in the lower right figure of , the electronic device 100 can reconstruct the second AI model without identifying the third product.

[0142] For example, when an image including the first product 701 to the third product 703 is input as input data, the reconstructed second AI model may output the first product 701 to the third product 703 as output data. However, when an image including the fourth product 704 is input as input data to the reconstructed second AI model, the electronic device 100 may output products other than the fourth product 704 as output data, or may output output data indicating that the image including the fourth product 704 is not identified. That is, in the case where the reliability of the output data associated with a specific image is less than a predetermined reliability, the electronic device 100 may recognize that the image including the fourth product 704 is an image associated with a new product. In this case, the electronic device 100 may transmit the image including the fourth product 704 to the external server 200, and obtain output data indicating that the image including the fourth product 704 corresponds to the fourth product 704.

[0143] For example, Figure 7BAs shown, in operation S710, the electronic device 100 may input input data to a second artificial intelligence (AI) model to obtain first output data. The input data may be an image including the fourth product 704, and the first output data may be a product other than the fourth product 704 (or an invalid message). In operation S720, the electronic device 100 may identify whether the first output data is less than or equal to a predetermined reliability value or a preset reliability value. That is, the electronic device 100 may identify the possibility that the first output data corresponds to a product other than the fourth product, or whether the first output data outputs an invalid message, and identify whether the first output data is less than or equal to a predetermined reliability value. In the case where the first output data is less than or equal to the predetermined reliability value or the preset reliability value, in operation S730, the electronic device 100 may transmit the input data (including the image of the fourth product 704) to the external server 200. In operation S740, the external server 200 may input the input data to the first artificial intelligence (AI) model to obtain second output data. The second output data may be data for the fourth product 704. In operation S750, the external server 200 may transmit the second output data to the electronic device 100. In operation S760, the electronic device 100 may acquire usage information based on the acquired second output data, and acquire reconstruction information based on the acquired usage information. In operation S770, the electronic device 100 may reconstruct the second AI model based on the acquired reconstruction information to acquire a third AI model.

[0144] exist Figure 7BIn the sequence diagram, in the case where the electronic device 100 does not obtain accurate output data for the image including the fourth product 704, the electronic device 100 obtains the second output data from the external server 200. However, the example is not limited to this. For example, in the case where the electronic device 100 does not obtain accurate output data for the image including the fourth product 704, the electronic device 100 may initialize the reconstructed second AI model and obtain the first AI model, input the image including the fourth product 704 to the obtained first AI model and obtain the second output data. That is, when the electronic device 100 is an electronic device 100 (such as a smart phone) that the user has used for a predetermined time or longer than a predetermined time period (for example, a long time such as six hours or longer in a day), and communication with the external server 200 is easy, the electronic device 100 may obtain the second output data from the external server. In the case where the electronic device 100 is an electronic device 100 such as a refrigerator, the user uses it for a predetermined time or less (for example, a short time such as one hour or less in a day) exceeding a predetermined time period, and communication with the external server 200 is not easy, the electronic device 100 may acquire the first AI model stored in the memory 110 to initialize the second AI model as the first AI model, and input input data to the acquired first AI model and acquire second output data.

[0145] exist Figure 7B In the sequence diagram, when the electronic device 100 does not obtain accurate output data for the image including the fourth product 704, the second AI model is immediately reconstructed into the third AI model. However, the example is not limited to this. That is, the electronic device 100 can reconstruct the second AI model into the third AI model only when the second output data satisfies a predetermined condition. For example, when the image including the fourth product 704 is input a predetermined number of times within a predetermined time, the electronic device 100 may reconstruct the second AI model into the third AI model. That is, when an image including the fourth product 704 that cannot be identified by the second AI model is transmitted to an external server and information about the fourth product 704 is obtained a predetermined number of times, the electronic device 100 may reconstruct the second AI model into the third AI model.

[0146] like Fig. 7AAs shown in the lower right figure of , if any one or any combination of the first product 701 to the third product 703 (for example, the third product 703) does not meet the predetermined condition, the electronic device 100 may obtain reconstruction information based on product information other than the third product, and reconstruct the second AI model based on the obtained reconstruction information to obtain a fourth artificial intelligence (AI) model. For example, if the output number of the third product 703 within a predetermined time period is less than a threshold, the electronic device 100 may obtain reconstruction information based on the first product 701 and the second product 702, and reconstruct the second AI model to obtain a fourth AI model.

[0147] In the above embodiment, the usage information is obtained based on the output data, but it is not limited thereto. That is, the usage information can be directly input through user settings. For example, Fig. 10A As shown, the electronic device 100 may display a user interface (UI) for adding a new product. When a user command is input through the UI for displaying a new product, the electronic device 100 may acquire usage information including a product corresponding to the input user command and a product corresponding to the output data, and acquire reconstruction information corresponding to the acquired usage information. Fig. 10A In the present invention, the electronic device 100 is described as a smart phone for convenience of explanation. However, the electronic device 100 may be various types of electronic devices including a display, and even without a display, the electronic device 100 obtains input of a user command through voice recognition and a remote control device.

[0148] exist Fig. 7A and Figure 7B In the embodiment, a method of reconstructing an artificial intelligence (AI) model for image analysis is described in the case where the electronic device 100 is a refrigerator, but it is not limited thereto. That is, the technical concept of the present disclosure can be applied to various situations in which image analysis is used in the electronic device 100.

[0149] Fig. 8A is a diagram providing an example for explaining a method for reconstructing a language translation artificial intelligence (AI) model according to another embodiment of the present disclosure. Figure 8B is a diagram providing an example for explaining a method for reconstructing a language translation artificial intelligence (AI) model according to another embodiment of the present disclosure.

[0150] The first AI model may be a language translation model when the user voice or text input is in a first language, for translating the input voice or input text into a second language and outputting the input voice or text in the second language. In this case, the first AI model stores a large number of words to translate various contexts and expressions. For example, the first AI model may store 10,000 words or more and handle language translation.

[0151] However, a single user using the electronic device 100 usually performs language translation only in a specific context, or more often, the word expression used is predetermined. Therefore, the electronic device 100 may obtain reconstruction information based on the usage information, reconstruct the first AI model based on the obtained reconstruction information, and obtain the second AI model. In this case, the usage information may include at least one of the user's context information, location information, schedule information, or expression. The user's context information may include information about the environment in which the user is to perform the translation, the user's age information, the user's occupation information, the user's attention information, etc. The information about the environment in which the user is to perform the translation may, for example, represent translation information related to words used in various environments, such as translation information related to words used in travel, translation information related to words used in hospitals, translation information related to words used in restaurants, translation information related to words used in business conversations, etc. The user's location information is information about the location of the user. According to the user's location information, environmental information (when the user's location information is a pharmacy, hospital, restaurant, etc.), the language to be translated or the language expression (such as an expression suitable for a high-end restaurant and an expression related to a rural accent).

[0152] Some of the various usage information described above may be obtained from output data output by an artificial intelligence (AI) model. For example, at least some of the user's contextual information, location information, schedule information, or expressions may be obtained from speech or text associated with a second language, the second language being output from speech or text associated with a first language input to the AI ​​model by the user.

[0153] However, in some cases, some of the above-mentioned various usage information may not be obtained from the output data output from the artificial intelligence (AI) model. For example, the user's context information, location information, schedule information, etc. are more likely to be obtained through user settings rather than from the output data output from the AI ​​model.

[0154] Fig. 10A is a diagram illustrating an example user interface (UI) screen for inputting usage information. Fig. 10B is a diagram illustrating an example user interface (UI) screen for inputting usage information.

[0155] For example, Fig. 10A As shown, the electronic device 100 may set the usage information according to a user command such as inputting “yes” in response to a question “Do you want to add a new product?” Fig. 10B A user interface (UI) screen capable of inputting the user's usage information or context information is shown. However, the example is not limited thereto, and schedule information, location information, etc. may be added to the usage information according to the user's selection. Alternatively, in the case where a predetermined condition is satisfied, even if the user is about Fig. 10B If the user command does not exist, some usage information that cannot be obtained from the output data output by the self-AI model can also be obtained as the usage information. For example, when a user command to use the AI ​​model is input, the electronic device 100 can obtain the location information of the electronic device 100 according to the input user command, and obtain the obtained location information as the usage information.

[0156] Alternatively, the electronic device 100 may acquire usage information based on sensing information acquired through various sensors. For example, the electronic device 100 may acquire usage information based on location information acquired via GPS, images acquired via image sensors, and the like.

[0157] The electronic device 100 may obtain reconstruction information based on the above-mentioned various context information. The reconstruction information may be information for reconstructing the AI ​​model with a focus on words corresponding to the usage information. For example, in the case where the usage information includes environmental information about a pharmacy, the reconstruction information may be information including words used in the pharmacy, words related to medicines, and the like.

[0158] The electronic device 100 may reconstruct the first AI model into the second AI model by using the reconstruction information. Figure 8B As shown, the electronic device 100 may reconstruct the first AI model 810 to obtain the second AI model 820 by adding rows corresponding to words corresponding to the usage information to the word embedding layer included in the first AI model 810, deleting rows corresponding to words irrelevant to the usage information, changing the positions of rows corresponding to words corresponding to the usage information, etc. Alternatively, the electronic device 100 may reconstruct the first AI model 810 to obtain the second AI model 820, and change the weight values ​​of the first AI model 810 by using the reconstruction information.

[0159] Fig.9A are drawings provided to explain various embodiments of the present disclosure. Fig. 9B are drawings provided to explain various embodiments of the present disclosure. Fig. 9C It is provided to explain various embodiments of the present disclosure.

[0160] like Fig.9A , Fig. 9B and Fig. 9C As shown, the electronic device 100 may be, for example, an artificial intelligence (AI) speaker. However, the example is not limited thereto, and the electronic device 100 may be of various types using a corresponding AI model, such as a smart phone, a navigation device, etc.

[0161] Fig.9A is a diagram provided for explaining an example method for reconstructing an artificial intelligence (AI) model to update words frequently used by a user through speech recognition.

[0162] The electronic device 100 may input a user voice into a first artificial intelligence (AI) model and output a response to the user voice. The electronic device 100 may use the first AI model to obtain data about various words included in the user voice. When the obtained word data satisfies a predetermined condition, the electronic device 100 may obtain usage information based on the word data that satisfies the predetermined condition, and reconstruct the first AI model based on the obtained usage information and obtain a second AI model. The predetermined condition may, for example, be data for a word obtained a predetermined number of times or more within a predetermined time period. As shown in FIG. Fig. 7A As described, as user voice data accumulates, new word data may be added or deleted, and the electronic device 100 may reconstruct the reconstructed second AI model into a third artificial intelligence (AI) model or a fourth artificial intelligence (AI) model through the above-mentioned various methods.

[0163] Fig. 9B is a diagram provided for explaining an example method for reconstructing an artificial intelligence (AI) model for removing noise.

[0164] The electronic device 100 may input a user voice into the first AI model as input data, and obtain information about the location where the user voice is input as output data. The electronic device 100 may remove noise input into the electronic device 100 based on the obtained location information. In this case, the electronic device 100 may obtain usage information based on the information about the location where the electronic device 100 is located output by the first AI model. The acquired usage information may be information related to the structure and size of the location and the location of the electronic device 100 within the location. As described above, in the case where the usage information cannot be obtained from the output data, the electronic device 100 may obtain new usage information based on a user command or external information. The electronic device 100 may obtain reconstruction information based on the acquired usage information, and may reconstruct the first AI model based on the acquired reconstruction information and obtain a second AI model. In the case where the location of the electronic device 100 is changed, the electronic device 100 may reconstruct the AI ​​model according to the new location. Fig. 9B In the present invention, a method for removing noise from a user's voice is provided. However, the example is not limited thereto, and the technical idea of ​​the present disclosure can be applied to a case where data sensed by a sensor (eg, a camera, a microphone, etc.) included in the electronic device 100 is processed.

[0165] Fig. 9C is a diagram provided for explaining an example method for reconstructing an artificial intelligence (AI) model for user recognition.

[0166] The electronic device 100 may input user voices associated with multiple users into a first AI model, and output user identification information corresponding to the user voices. In the case where the user identification information corresponding to the user voices satisfies a predetermined condition, the electronic device 100 may obtain the user identification information that satisfies the predetermined condition as usage information, reconstruct the first AI model based on the obtained usage information, and obtain the second AI model. The predetermined condition may, for example, be obtaining user identification information a predetermined number of times or more within a predetermined time period. That is, the electronic device 100 may identify users who frequently use the electronic device 100, and reconstruct the AI ​​model to provide a mitigated AI model for the identified users. In addition, the electronic device 100 may reconstruct an artificial intelligence (AI) model for identifying user voices based on the user voices corresponding to the user identification information. That is, as referenced Fig. 7A As described, the electronic device 100 can reconstruct an AI model for speech recognition.

[0167] Fig.11 is a block diagram provided to explain the operation of an electronic device using an artificial intelligence (AI) model according to an embodiment of the present disclosure.

[0168] The processor 120 may include a data learning component 1110 and a data identification component 1120. The data learning component 1110 may train an artificial intelligence (AI) model so that the AI ​​model includes criteria according to a specific purpose. The specific purpose may include data identification related to speech recognition, translation, image recognition, context recognition, etc. In addition, the specific purpose may include purposes related to classification, grouping, and clustering of data. In order to identify operations according to the above purposes, the data learning component 1110 may apply learning data to the AI ​​model and generate an AI model including a determination standard. The data identification component 1120 may identify contexts about a specific purpose based on input data. The data identification component 1120 may use the trained AI model to identify contexts from predetermined recognition data. The data identification component 1120 may obtain predetermined input data according to a predetermined standard, apply the obtained input data as input data to the AI ​​model, and thereby identify (or estimate) a predetermined context based on the predetermined input data. In addition, the result value output by applying the obtained input data to the AI ​​model as an input value may be used to update the AI ​​model.

[0169] At least some of the data learning component 1110 and the data identification component 1120 may be implemented as a software module or manufactured as at least one hardware chip and installed on an electronic device. For example, one or both of the data learning component 1110 and the data identification component 1120 may be manufactured as a hardware chip for artificial intelligence (AI), or manufactured as a part of an existing general-purpose processor (e.g., a CPU or an application processor) or a graphics-specific processor (e.g., a GPU), and installed on the above-mentioned various electronic devices. In this case, the AI-specific hardware chip is a dedicated processor dedicated to probability calculation, which has a higher parallel processing performance than the existing general-purpose processor, so it can quickly process computing operations in the field of artificial intelligence such as machine learning. In the case where the data learning component 1110 and the data identification component 1120 are implemented as a software module (or a program module including instructions), the software module may be stored in a non-temporary computer-readable medium. The software module may be provided by an operating system (OS) or a predetermined application. Alternatively, a portion of the software module may be provided by the OS, and the rest may be provided by a predetermined application.

[0170] In this case, the data learning component 1110 and the data identification component 1120 may be installed on one electronic device 100, or may be installed on separate electronic devices respectively. For example, the electronic device 100 may include only the data learning component 1110 for training the AI ​​model, and the external device may include only the data identification component 1120. Alternatively, the electronic device 100 may include the data learning component 1110 and the data identification component 1120, and the external electronic device may include only the data identification component 1120. The data learning component 1110 and the data identification component 1120 may provide the model information constructed by the data learning component 1110 to the data identification component 1120 via a cable or wirelessly, and the data input to the data identification component 1120 may be provided to the data identification component 1120 as additional learning data. As another example, the data learning component 1110 and the data identification component 1120 may be elements of the external server 200. In this case, when the electronic device 100 transmits the learning data or the input data to the external server 200 , the external server 200 may transmit a result value associated with the learning data or the input data acquired from the electronic device 100 to the electronic device 100 .

[0171] The data learning component 1110 may also include a data acquisition component, a preprocessing component, a learning data selection component, a model learning component, and a model evaluation component. The data acquisition component may be configured to acquire learning data according to a specific purpose. The preprocessing component is a component for preprocessing the data acquired from the acquisition component into a predetermined format. The learning data selection component is a component for selecting data for learning from the data acquired by the learning data acquisition component and the data preprocessed by the learning data preprocessing component. The selected learning data may be provided to the model learning component. The model learning component is a component for training an artificial intelligence (AI) model by using the learning data. The model evaluation component is a component for improving the results of the AI ​​model.

[0172] Any one or any combination of the above-mentioned data acquisition components, preprocessing components, learning data selection components, model learning components, and model evaluation components can be implemented as a software module or manufactured as at least one hardware chip and installed on an electronic device. For example, any one or any combination of the data learning component, preprocessing component, learning data selection component, model learning component, and model evaluation component can be manufactured as a hardware chip for artificial intelligence (AI), or manufactured as part of an existing general-purpose processor (e.g., a CPU or an application processor) or a graphics-specific processor (e.g., a GPU), and installed on the above-mentioned various electronic devices.

[0173] In addition, the data identification component 1120 may also include a data acquisition component, a preprocessing component, an input data selection component, a recognition result providing component, and a model updating component. The data acquisition component is a component for acquiring input data. The preprocessing component is a component for preprocessing the data acquired from the acquisition component into a predetermined format. The input data selection component is a component for selecting data for recognition from the preprocessed data. The recognition result providing component is a component capable of acquiring data selected from the input data. The model updating component is a component for updating an artificial intelligence (AI) model based on the evaluation of the recognition result provided by the recognition result providing component.

[0174] Any one or any combination of the above-mentioned data acquisition component, input data selection component, recognition result providing component and model updating component can be implemented as a software module or manufactured as at least one hardware chip and installed on an electronic device. For example, any one or any combination of the data learning component, preprocessing component, learning data selection component, model learning component and model evaluation component can be manufactured as a hardware chip for artificial intelligence (AI), or manufactured as a part of an existing general-purpose processor (e.g., CPU or application processor) or a graphics-specific processor (e.g., GPU), and installed on the above-mentioned various electronic devices.

[0175] Fig.12is a flowchart provided to explain a control method of an electronic device according to an embodiment of the present disclosure.

[0176] In operation S1210, the electronic device 100 may obtain at least one output data associated with at least one input data input to a first artificial intelligence (AI) model. As described above, depending on the type of AI model, the input data and the output data may be different. For example, in the case where the AI ​​model is an image analysis model, the input data may be image data, and the output data may be information about an object included in the image data. As another example, in the case where the AI ​​model is a language translation model, the input data may be speech or text in a first language, and the output data may be speech or text in a second language for which the speech or text in the first language is to be translated. As another example, in the case where the AI ​​model is a speech recognition model, the input data may be user speech, and the output data may be a response to the user speech.

[0177] In operation S1220, the electronic device 100 may acquire first usage information based on the acquired output data. As described above, the usage information may include any one or any combination of information about the user's usage pattern, information about the location of the electronic device 100, age information of the user of the electronic device, context information, and environmental information.

[0178] In operation S1230, the electronic device 100 may acquire first reconstruction information for reconstructing the first AI model based on the acquired first usage information. As described above, the reconstruction information may include any one or any combination of information about the structure of the AI ​​model, information about the weight of the AI ​​model, and incremental information for reconstructing the first AI model.

[0179] In operation S1240, the electronic device 100 may reconstruct the first AI model based on the first reconstruction information to obtain a second AI model. The reconstructed second AI model may be reconstructed multiple times as needed into a third artificial intelligence (AI) model or a fourth artificial intelligence (AI) model.

[0180] The term "component" or "module" used herein includes a unit composed of hardware, software or firmware, and can be used interchangeably with terms such as logic, logic block, component or circuit. A "component" or "module" can be a component of an overall structure that performs one or more functions or its smallest unit or part. For example, a module may include an application-specific integrated circuit (ASIC).

[0181] The above-mentioned various embodiments can be implemented as an S / W program including instructions stored on a machine-readable (e.g., computer-readable) storage medium. A machine is a device that can call stored instructions from a storage medium and operate according to the called instructions, and the device may include an electronic device (e.g., electronic device 100) according to the above-mentioned embodiment. When the instruction is executed by a processor, the processor may perform a function corresponding to the instruction directly or by other components under the control of the processor. One or more instructions may include code generated by a compiler or code executable by an interpreter. The machine-readable storage medium may be provided as a non-transitory storage medium. In this article, the term "non-transitory" only means that the storage medium does not include a signal but is tangible, and does not distinguish between semi-permanent data stored in the storage medium and data temporarily stored in the storage medium.

[0182] According to one embodiment, the method according to each of the above embodiments may be configured to be included in a computer program product. The computer program product may be traded as a product between a seller and a consumer. The computer program product may be distributed as a machine-readable storage medium (e.g., a compact disc read-only memory (CD-ROM)), or may be available through an online application store (e.g., Play Store TM ) Online distribution. For online distribution, at least a part of the computer program product may be at least temporarily stored or retrieved on a storage medium such as a memory of a manufacturer's server, an application store's server or a relay server.

[0183] According to each embodiment, each element (e.g., module or program) can be configured as a single or multiple entity, and some corresponding sub-elements mentioned above can be omitted, or another sub-element can be included in each embodiment. Alternatively or additionally, some elements (e.g., modules or programs) can be integrated into one entity, and the functions performed by each element before integration can be performed in the same or similar manner. According to each embodiment, the module, program or operation run by other elements can be performed continuously, in parallel, repeatedly or tentatively, or at least some operations can be run according to different orders, and other operations can be omitted or added thereto.

Claims

1. A control method for an electronic device, the control method comprising: Inputting at least one input data into a first artificial intelligence (AI) model to obtain at least one output data output by the first AI model; acquiring first usage information based on the acquired at least one output data; Determining whether the acquired first usage information satisfies a predetermined condition, wherein the predetermined condition includes that a number of times that the first data of the at least one output data is output by the first AI model is greater than or equal to a threshold; In response to determining that the acquired first usage information satisfies a predetermined condition, wherein the number of times the first data of the at least one output data is output by the first AI model is greater than or equal to a threshold: Based on the acquired first usage information, acquiring first reconstruction information for reconstructing the first AI model; and Reconstruct the first AI model based on the acquired first reconstruction information to obtain a second AI model; In response to determining that the first usage information does not satisfy a predetermined condition, wherein the number of times that the first data of the at least one output data is output by the first AI model is less than a threshold: Acquire second usage information based on the at least one output data acquired, the second usage information not including first data outputted by the first AI model less than a threshold value in the at least one output data; acquiring second reconstruction information for reconstructing the first AI model based on the acquired second usage information; and The first AI model is reconstructed based on the acquired second reconstruction information to obtain a third AI model.

2. The control method according to claim 1, wherein: The first usage information includes any one or any combination of information related to a user's usage pattern, information related to a location where the electronic device is located, age information of a user of the electronic device, context information, and environmental information.

3. The control method according to claim 1, wherein: The first reconstruction information includes any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model, and incremental information for reconstructing the first AI model, and The incremental information includes one or both of structural information of the first AI model and / or the second AI model and information about a weight difference between the first AI model and the second AI model.

4. The control method according to claim 1, further comprising: Inputting the first input data into the obtained second AI model to obtain second output data; transmitting the first input data to the server based on the reliability of the acquired second output data being less than or equal to a predetermined value; as well as Third output data associated with the transmitted first input data is obtained from the server.

5. The control method according to claim 4, further comprising: acquiring third usage information based on the acquired third output data; Based on the acquired third usage information, acquire third reconstruction information for reconstructing the second AI model; as well as Based on the acquired third reconstruction information, the acquired second AI model is reconstructed to acquire a fourth AI model.

6. The control method according to claim 1, further comprising: displaying a user interface (UI) for adding and / or modifying the acquired first usage information; modifying the acquired first usage information based on a user command for adding and / or changing the acquired first usage information being input through the displayed UI; Based on the modified first usage information, obtaining third reconstruction information for reconstructing the second AI model; and Based on the acquired third reconstruction information, the acquired second AI model is reconstructed to acquire a fourth AI model.

7. The control method according to claim 1, wherein: Each of the first AI model and the second AI model is used for image analysis or for voice processing and language understanding.

8. An electronic device comprising: a memory for storing instructions; as well as A processor configured to execute the stored instructions to: Inputting at least one input data into a first artificial intelligence (AI) model to obtain at least one output data output by the first AI model; acquiring first usage information based on the acquired at least one output data; Determining whether the acquired first usage information satisfies a predetermined condition, wherein the predetermined condition includes that a number of times that the first data of the at least one output data is output by the first AI model is greater than or equal to a threshold; In response to determining that the acquired first usage information satisfies a predetermined condition, wherein the number of times the first data of the at least one output data is output by the first AI model is greater than or equal to a threshold: Based on the acquired first usage information, acquiring first reconstruction information for reconstructing the first AI model; and Reconstruct the first AI model based on the acquired first reconstruction information to obtain a second AI model; In response to determining that the first usage information does not satisfy a predetermined condition, wherein the number of times that the first data of the at least one output data is output by the first AI model is less than a threshold: Acquire second usage information based on the at least one output data acquired, the second usage information not including first data outputted by the first AI model less than a threshold value in the at least one output data; acquiring second reconstruction information for reconstructing the first AI model based on the acquired second usage information; and The first AI model is reconstructed based on the acquired second reconstruction information to obtain a third AI model.

9. The electronic device according to claim 8, wherein: The first usage information includes any one or any combination of information related to a user's usage pattern, information related to a location where the electronic device is located, age information of a user of the electronic device, context information, and environmental information.

10. The electronic device according to claim 8, wherein: The first reconstruction information includes any one or any combination of information about the structure of the second AI model, information about the weight of the second AI model, and incremental information for reconstructing the first AI model, and The incremental information includes one or both of structural information of the first AI model and / or the second AI model and information about a weight difference between the first AI model and the second AI model.

11. The electronic device according to claim 8, further comprising: a communicator, the communicator comprising a circuit; Wherein, the processor is further configured to execute the stored instructions to: Inputting the first input data into the obtained second AI model to obtain second output data; Based on the reliability of the acquired second output data being less than or equal to a predetermined value, controlling the communicator to transmit the first input data to the server via the communicator; and Third output data associated with the transmitted first input data is acquired from the server via the communicator.

12. The electronic device according to claim 11, wherein: The processor is further configured to execute the stored instructions to: acquiring third usage information based on the acquired third output data; Based on the acquired third usage information, acquiring third reconstruction information for reconstructing the second AI model; and Based on the acquired third reconstruction information, the acquired second AI model is reconstructed to acquire a fourth AI model.

13. The electronic device according to claim 8, wherein: Each of the first AI model and the second AI model is used for image analysis or for voice processing and language understanding.

14. The electronic device according to claim 8, further comprising a display, in, The processor is further configured to execute the stored instructions to: controlling the display to display a user interface (UI) for adding and / or modifying the acquired first usage information; modifying the acquired first usage information based on a user command for adding and / or changing the acquired first usage information being input through the displayed UI; Based on the modified first usage information, obtaining third reconstruction information for reconstructing the second AI model; as well as Based on the acquired third reconstruction information, the acquired second AI model is reconstructed to acquire a fourth AI model.

Images