Electronic device and compensation method thereof

Abstract

An electronic device is disclosed. The electronic device comprises: a sensor; a memory in which an artificial intelligence model is stored; and one or more processors which obtain a compensation value for compensating for data drift related to the artificial intelligence model when it is identified that the data drift has occurred on the basis of an input value of the artificial intelligence model obtained by using the sensor, and perform compensation for the data drift on the basis of the compensation value.

Description

Electronic devices and their compensation methods

[0001] The present disclosure relates to an electronic device for compensating for data drift and a compensation method thereof.

[0002] Recently, advancements in artificial intelligence-related technologies have continued, and in particular, electronic devices storing artificial intelligence models have recently become available.

[0003] These on-device AI (artificial intelligence) products are generally vulnerable to data drift that can occur due to component aging, as software updates are not easy.

[0004] An electronic device according to an embodiment of the present disclosure includes a sensor, a memory storing an artificial intelligence model, and one or more processors. If the one or more processors determine that data drift has occurred in the artificial intelligence model based on input values ​​of the artificial intelligence model acquired using the sensor, the one or more processors acquire a compensation value to compensate for the data drift. The one or more processors compensate for the data drift based on the compensation value.

[0005] Additionally, the one or more processors may obtain a first compensation value for compensating an input value of the artificial intelligence model, and input the compensated input value based on the first compensation value into the artificial intelligence model to obtain a predicted value of the artificial intelligence model from the artificial intelligence model.

[0006] In addition, the one or more processors may obtain a second compensation value for compensating the predicted value of the artificial intelligence model, obtain the predicted value of the artificial intelligence model obtained by inputting the input value into the artificial intelligence model, and compensate the predicted value based on the second compensation value.

[0007] Additionally, the memory may store information regarding a plurality of compensation values ​​corresponding to a plurality of accumulated usage times. The one or more processors may obtain a compensation value corresponding to the accumulated usage time of the electronic device from among the plurality of compensation values ​​corresponding to the plurality of accumulated usage times.

[0008] Additionally, the one or more processors can obtain the compensation value using the predicted value of the artificial intelligence model and the actual value detected using the sensor.

[0009] In addition, the one or more processors may obtain a plurality of input values ​​by applying each of the plurality of candidate reward values ​​to an input value of the artificial intelligence model, obtain a plurality of predicted values ​​of the artificial intelligence model by inputting each of the plurality of input values ​​into the artificial intelligence model, identify a prediction performance of the artificial intelligence model corresponding to each of the plurality of candidate reward values ​​based on each of the plurality of predicted values ​​and an actual value detected using the sensor, and determine the reward value from among the plurality of candidate reward values ​​based on the identified prediction performance.

[0010] Additionally, the one or more processors can identify whether the data drift has occurred based on the learning value of the artificial intelligence model and the input value of the artificial intelligence model obtained using the sensor.

[0011] A method for compensating for data drift of an electronic device storing an artificial intelligence model according to an embodiment of the present disclosure includes, when it is determined that data drift has occurred for the artificial intelligence model based on input values ​​of the artificial intelligence model acquired using a sensor, a step of obtaining a compensation value for compensating for the data drift, and a step of performing compensation for the data drift based on the compensation value.

[0012] In a non-transitory computer-readable medium storing computer instructions that cause an electronic device to perform an operation when executed by one or more processors of an electronic device storing an artificial intelligence model according to an embodiment of the present disclosure, the operation includes a step of obtaining a compensation value for compensating for the data drift when it is determined that a data drift has occurred with respect to the artificial intelligence model based on an input value of the artificial intelligence model acquired using a sensor, and a step of performing compensation for the data drift based on the compensation value.

[0013] FIG. 1A is a block diagram illustrating the configuration of an electronic device according to an embodiment of the present disclosure.

[0014] FIG. 1b is a block diagram illustrating the configuration of an electronic device according to an embodiment of the present disclosure.

[0015] FIG. 2 is a flowchart illustrating an operation of an electronic device to identify whether data drift has occurred according to an embodiment of the present disclosure.

[0016] FIG. 3 is a flowchart illustrating an operation of an electronic device performing compensation for data drift according to an embodiment of the present disclosure.

[0017] FIG. 4A and FIG. 4B are drawings for explaining an example of a compensation value for compensating an input value of an artificial intelligence model according to an embodiment of the present disclosure.

[0018] FIG. 5 is a diagram illustrating an example of a compensation value for compensating a predicted value of an artificial intelligence model according to an embodiment of the present disclosure.

[0019] FIG. 6 is a diagram showing an example of a value corresponding to the prediction performance of an artificial intelligence model obtained based on a plurality of candidate reward values ​​according to an embodiment of the present disclosure.

[0020] FIG. 7 is a flowchart illustrating a compensation method of an electronic device according to an embodiment of the present disclosure.

[0021] Hereinafter, terms used in this specification will be briefly described, and the present disclosure will be described in detail. In this disclosure, the expression “at least one of a, b, or c” can refer to “a,” “b,” “c,” “a and b,” “a and c,” “b and c,” “all of a, b, and c,” or variations thereof.

[0022] The terms used in this disclosure are selected from widely used, common terms, taking into account the functions of the disclosure. However, these terms may vary depending on the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in certain cases, terms may be arbitrarily selected by the applicant, in which case their meanings will be described in detail in the relevant description. Therefore, the terms used in this disclosure should not be defined simply as names, but rather based on the meanings of the terms and the overall content of the disclosure.

[0023] Singular expressions may include plural expressions unless the context clearly indicates otherwise. Terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art described herein. Furthermore, terms containing ordinal numbers, such as "first" or "second," used herein may be used to describe various components, but such components should not be limited by such terms. Such terms are used solely to distinguish one component from another.

[0024] When a part of the specification is said to "include" a component, unless otherwise specifically stated, this does not exclude other components but rather implies the inclusion of other components. Furthermore, terms such as "part" and "module" used in the specification refer to a unit that processes at least one function or operation, which may be implemented in hardware, software, or a combination of hardware and software.

[0025] The term "and / or" includes any combination of a plurality of related described elements or any one of a plurality of related described elements.

[0026] Meanwhile, the various elements and areas in the drawings are schematically drawn. Therefore, the technical concept of the present invention is not limited by the relative sizes or spacing depicted in the attached drawings.

[0027] The present disclosure will be described below with reference to the attached drawings.

[0028] FIG. 1A is a block diagram illustrating the configuration of an electronic device according to an embodiment of the present disclosure.

[0029] Referring to FIG. 1A, the electronic device (100) may include a memory (110), a sensor (120), and one or more processors (130). The configuration of the electronic device (100) illustrated in FIG. 1A is merely an example, and it is to be understood that additional configurations may be added depending on the embodiment. For example, the electronic device (100) may be various types of electronic devices, such as a television, a refrigerator, an air conditioner, a washing machine, a robot vacuum cleaner, and the like.

[0030] Memory (110) may store instructions, data structures, and program codes. Operations performed by one or more processors (130) may be implemented by executing instructions or codes of a program stored in memory (110).

[0031] The memory (110) may include a flash memory type, a hard disk type, a multimedia card micro type, a card type memory (e.g., SD or XD memory, etc.), and may include a non-volatile memory including at least one of a ROM (Read-Only Memory), an EEPROM (Electrically Erasable Programmable Read-Only Memory), a PROM (Programmable Read-Only Memory), a magnetic memory, a magnetic disk, and an optical disk, and a volatile memory such as a RAM (Random Access Memory) or an SRAM (Static Random Access Memory).

[0032] The memory (110) may store an artificial intelligence model (111). The artificial intelligence model (111) may be an artificial intelligence model trained to predict information related to the electronic device (100). The prediction may be replaced with an expression such as an inference, for example. In addition, the predicted value of the artificial intelligence model (111) may be replaced with an expression such as predicted data, an output value, or output data.

[0033] Additionally, learning values ​​used for learning an artificial intelligence model (111) may be stored in the memory (110). The learning values ​​may be replaced with expressions such as learning data, for example.

[0034] The information predicted by the artificial intelligence model (111) may include various information about the electronic device (100).

[0035] For example, if the electronic device (100) is a refrigerator (100), the artificial intelligence model (111) may include a neural network model trained to predict the temperature of a storage compartment after a certain period of time. For example, if the electronic device (100) is a robot vacuum cleaner (100), the artificial intelligence model (111) may include a neural network model trained to predict whether the robot vacuum cleaner (100) is driving on a carpet. However, the present invention is not limited thereto, and the artificial intelligence model (111) may predict various information about the electronic device (100) depending on the type of the electronic device (100).

[0036] The neural network model according to the present disclosure refers to an artificial intelligence model including a neural network and can be trained by deep learning. The neural network may include, for example, at least one of a convolutional neural network (CNN), a recurrent neural network (RNN), a restricted boltzmann machine (RBM), a deep belief network (DBN), a bidirectional recurrent deep neural network (BRDNN), a generative adversarial network (GAN), and a deep Q-network. However, the neural network model is not limited to the examples described above.

[0037] In this way, the artificial intelligence model (111) can be implemented in the form of an on-device.

[0038] The sensor (120) is a configuration for detecting data regarding the electronic device (100). One or more processors (130) can obtain data regarding the electronic device (100) using the sensor (120).

[0039] Data detected by the sensor (120) may include data regarding the status of the electronic device (100) and data regarding the status of components (e.g., parts, batteries, etc.) included in the electronic device (100).

[0040] In this case, the data detected by the sensor (120) may include an output value of a component of the electronic device (100). The output value of the component may include, for example, an output value of a motor (e.g., rotational speed, etc.), an output value of an IMU (Inertial Measurement Unit) sensor (e.g., acceleration value, angular velocity value, etc.), an output voltage of a battery, etc. The output value of such a component may be used as an input value of an artificial intelligence model (111). The input value may be replaced with an expression such as input data, for example.

[0041] For example, assume that the electronic device (100) is a refrigerator (100). The refrigerator (100) may include a cold air supply device that supplies cold air to a storage compartment. The cold air supply device may generate cold air through a refrigeration cycle including a process of compression, condensation, expansion, and evaporation of a refrigerant. To this end, the cold air supply device may include a refrigeration cycle device having a compressor, a condenser, an expansion device, and an evaporator, and a fan for supplying the generated cold air to the storage compartment. In this case, the sensor (120) may include a sensor (e.g., an encoder) for detecting the rotation speed of the motor. The refrigerator (100) inputs information detected by the sensor (120), that is, the rotation speed of the motor driving the fan (e.g., a fan motor) and the rotation speed of the motor driving the compressor (e.g., a compressor motor), into an artificial intelligence model (111) trained to predict the temperature of the storage compartment after a certain period of time, thereby obtaining information on the temperature of the storage compartment after a certain period of time predicted by the artificial intelligence model (111).

[0042] For example, it is assumed that the electronic device (100) is a robot cleaner (100). The sensor (120) may include a sensor (e.g., an IMU sensor, etc.) for detecting the movement of the electronic device (100). The robot cleaner (100) inputs information detected by the sensor (120), i.e., an acceleration value and an angular velocity value of the robot cleaner (100), into an artificial intelligence model (111) trained to identify whether the robot cleaner (100) is driving on a carpet, obtains a probability value from the artificial intelligence model (111), and compares the probability value with a threshold value to identify whether the robot cleaner (100) is driving on a carpet.

[0043] However, it is not limited thereto, and the sensor (120) can detect various data used as input values ​​of the artificial intelligence model (111) depending on the type of the artificial intelligence model (111).

[0044] Additionally, data detected by the sensor (120) can be used to evaluate the predictive performance of the artificial intelligence model (111). Predictive performance can indicate how closely the predicted values ​​from the artificial intelligence model match the actual values. Predictive performance can be expressed as a value using methods such as the Root Mean Squared Error (RMSE) and the F1 score.

[0045] To this end, the data detected by the sensor (120) may include an actual value corresponding to a value predicted by the artificial intelligence model (111). For example, assume that the electronic device (100) is a refrigerator (100) and the artificial intelligence model (111) is trained to predict the temperature of the storage compartment after a certain period of time. In this case, the sensor (120) may include a temperature sensor for detecting the temperature of the storage compartment. One or more processors (130) may use the temperature sensor to measure the actual temperature of the storage compartment after a certain period of time, and compare the measured temperature with the temperature predicted by the artificial intelligence model (111) to identify the prediction performance of the artificial intelligence model (111).

[0046] However, it is not limited thereto, and the sensor (120) can detect various actual data corresponding to the values ​​predicted by the artificial intelligence model (111) depending on the type of the artificial intelligence model (111).

[0047] One or more processors (130) can control the overall operations of the electronic device (100). For example, the one or more processors (130) can identify whether data drift has occurred in the artificial intelligence model (111) of the electronic device (100) by executing one or more instructions stored in the memory (110), and control the overall operations to compensate for the data drift.

[0048] The one or more processors (130) may include one or more of a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), an Accelerated Processing Unit (APU), a Many Integrated Core (MIC), a Digital Signal Processor (DSP), a Neural Processing Unit (NPU), a hardware accelerator, or a machine learning accelerator. The one or more processors (130) may control one or any combination of other components of the electronic device (100) and may perform operations related to communication or data processing. The one or more processors (130) may execute one or more programs or instructions stored in the memory (110). For example, the one or more processors (130) may perform a method according to an embodiment of the present disclosure by executing one or more instructions stored in the memory (110).

[0049] When a method according to an embodiment of the present disclosure includes multiple operations, the multiple operations may be performed by one processor or by multiple processors. For example, when a first operation, a second operation, and a third operation are performed by a method according to an embodiment, the first operation, the second operation, and the third operation may all be performed by the first processor, or the first operation and the second operation may be performed by the first processor (e.g., a general-purpose processor) and the third operation may be performed by the second processor (e.g., an artificial intelligence-specific processor).

[0050] One or more processors (130) may be implemented as a single core processor including one core, or may be implemented as one or more multicore processors including multiple cores (e.g., homogeneous multicores or heterogeneous multicores). When one or more processors (130) are implemented as a multicore processor, each of the multiple cores included in the multicore processor may include an internal processor memory, such as a cache memory or an on-chip memory, and a common cache shared by the multiple cores may be included in the multicore processor. In addition, each of the multiple cores (or some of the multiple cores) included in the multicore processor may independently read and execute a program instruction for implementing a method according to an embodiment of the present disclosure, or all (or some) of the multiple cores may be linked to read and execute a program instruction for implementing a method according to an embodiment of the present disclosure.

[0051] When a method according to an embodiment of the present disclosure includes a plurality of operations, the plurality of operations may be performed by one core among the plurality of cores included in a multi-core processor, or may be performed by the plurality of cores. For example, when a first operation, a second operation, and a third operation are performed by a method according to an embodiment, the first operation, the second operation, and the third operation may all be performed by a first core included in the multi-core processor, or the first operation and the second operation may be performed by a first core included in the multi-core processor, and the third operation may be performed by a second core included in the multi-core processor.

[0052] In embodiments of the present disclosure, a processor may mean a system on a chip (SoC) in which one or more processors and other electronic components are integrated, a single-core processor, a multi-core processor, or a core included in a single-core processor or a multi-core processor, wherein the core may be implemented as a CPU, a GPU, an APU, a MIC, a DSP, an NPU, a hardware accelerator, or a machine learning accelerator, but embodiments of the present disclosure are not limited thereto.

[0053] FIG. 1b is a block diagram illustrating the configuration of an electronic device according to an embodiment of the present disclosure.

[0054] Referring to FIG. 1B, the electronic device (100) may include a memory (110), a sensor (120), one or more processors (130), a communication interface (140), an input interface (150), and an output interface (160). However, such a configuration is exemplary, and it is obvious that new configurations may be added or some configurations may be omitted in implementing the present disclosure. Meanwhile, among the configurations illustrated in FIG. 1B, a detailed description of configurations that overlap with the configuration illustrated in FIG. 1A will be omitted.

[0055] The communication interface (140) can perform data communication with external electronic devices under the control of one or more processors (130). The external electronic devices may include servers, home appliances, mobile devices (e.g., smartphones, tablet PCs, wearable devices, etc.).

[0056] For example, the communication interface (140) may include a communication circuit that can perform data communication between the electronic device (100) and an external electronic device using at least one of data communication methods including wired LAN, wireless LAN, Wi-Fi, Wi-Fi Direct, Bluetooth, ZigBee, Wi-Fi Direct (WFD), infrared Data Association (IrDA), Bluetooth Low Energy (BLE), Near Field Communication (NFC), Wireless Broadband Internet (Wibro), World Interoperability for Microwave Access (WiMAX), Shared Wireless Access Protocol (SWAP), Wireless Gigabit Alliances (WiGig), and RF communication.

[0057] The input interface (150) includes circuitry. The input interface (150) can receive user input and transmit the user input to one or more processors (130). For example, the input interface (150) can receive various user inputs for setting or selecting various functions supported by the electronic device (100).

[0058] The input interface (150) may include various types of input devices.

[0059] In one example, the input interface (150) may include a physical button. The physical button may include a function key or a dial button. The physical button may also be implemented as one or more keys.

[0060] In one example, the input interface (150) can receive user input using a touch method. For example, the input interface (150) can be implemented as a touch screen capable of performing the function of a display (161).

[0061] For example, the input interface (150) may receive a user's voice using a microphone. One or more processors (130) may perform a function corresponding to the user's voice using voice recognition. For example, one or more processors (130) may convert the user's voice into text data using a STT (Speech To Text) function, obtain control command data based on the text data, and perform a function corresponding to the user's voice based on the control command data. Depending on the embodiment, the STT function may be performed by an external server.

[0062] The output interface (160) may include a display (161) and a speaker (162).

[0063] The display (161) can display various screens. One or more processors (130) can display various notifications, messages, information, etc. related to the operation of the electronic device (100) on the display (161).

[0064] The display (161) may be implemented as a display including a self-luminous element or a display including a non-luminous element and a backlight. For example, the display (161) may be implemented as various types of displays such as an LCD (Liquid Crystal Display), an OLED (Organic Light Emitting Diodes) display, an LED (Light Emitting Diodes) display, a micro LED display, a Mini LED display, a QLED (Quantum dot light-emitting diodes) display, etc.

[0065] The speaker (162) can output audio signals. One or more processors (130) can output warning sounds, notification messages, response messages corresponding to user input, etc. related to the operation of the electronic device (100) through the speaker (162).

[0066] One or more processors (130) can control the operation of the electronic device (100) using the predicted values ​​of the artificial intelligence model (111).

[0067] For example, assume that the electronic device (100) is a refrigerator (100) and the artificial intelligence model (111) is a model trained to predict a certain period of time later. One or more processors (130) can obtain information about the temperature of the storage room after a certain period of time predicted by the artificial intelligence model (111) from the artificial intelligence model (111) and control the cooling air supply device based on the set temperature of the storage room and the predicted temperature.

[0068] For example, assume that the electronic device (100) is a robot vacuum cleaner (100) and the artificial intelligence model (111) is a model trained to identify whether the robot vacuum cleaner (100) is driving on a carpet. If the probability value obtained from the artificial intelligence model (111) is greater than or equal to a threshold value, one or more processors (130) can identify that the robot vacuum cleaner (100) is driving on a carpet and operate the cleaning device (e.g., a cleaning module for sucking up dust) of the robot vacuum cleaner (100) in a mode for cleaning the carpet.

[0069] However, it is not limited thereto, and the electronic device (100) can perform various operations according to the predicted value of the artificial intelligence model (111).

[0070] For convenience of explanation, one or more processors (130) are referred to as processors (130) below.

[0071] The processor (130) can identify whether a date drift has occurred in the artificial intelligence model based on the input value of the artificial intelligence model (111) obtained using the sensor (120).

[0072] Data drift can refer to differences in statistical characteristics between the training values ​​used to train an AI model and the input values ​​of the AI ​​model. Statistical characteristics can include the mean and variance.

[0073] For example, an on-device AI model can be trained using the normal output values ​​of components as learning values. The normal output values ​​may include the output values ​​of components according to the actual design of the electronic device (100) in a normal state where the electronic device (100) is not broken or aged.

[0074] However, when a component of an electronic device (100) ages, the performance of the component changes, and the output value of the component becomes different from the normal state. In this case, the change in the output value of the component due to aging may include an increase in the output value of the component compared to the normal state, or a decrease in the output value of the component compared to the normal state.

[0075] For example, if the component is a motor, the rotational speed of the motor may decrease with aging. Additionally, if the component is a battery, the output voltage of the battery may decrease with aging. However, the present disclosure is not limited thereto, and the output values ​​of various components may change with aging. For example, assume that the electronic device (100) is moved by the rotation of wheels (e.g., a robot vacuum cleaner). In this case, if the wheels wear out with aging, even if the electronic device (100) moves in the same environment, the movement of the electronic device (100) changes compared to before the wheels wear out, and accordingly, the output value of the IMU sensor may change from before.

[0076] When data drift like this occurs, the predictive performance of the AI ​​model deteriorates, requiring retraining. However, when the AI ​​model is implemented on-device, retraining and updating the AI ​​model may be difficult.

[0077] Accordingly, the processor (130) can identify whether data drift has occurred based on the input value of the artificial intelligence model and the learning value of the artificial intelligence model, and if it is identified that data drift has occurred, the data drift can be compensated for.

[0078] FIG. 2 is a flowchart illustrating an operation of an electronic device to identify whether data drift has occurred according to an embodiment of the present disclosure.

[0079] In operation S210, the processor (130) may obtain input values ​​of the artificial intelligence model (111). The input values ​​of the artificial intelligence model (111) may include data detected by the sensor (120). In this case, the processor (130) may collect input values ​​of the artificial intelligence model (111) using the sensor (120) for a certain period of time.

[0080] In operation S220, the processor (130) can identify the characteristics of the input values ​​of the artificial intelligence model (111) and the characteristics of the learning values ​​used for learning the artificial intelligence model (111). The learning values ​​of the artificial intelligence model (111) can be stored in the memory (110). The characteristics can include the mean and variance. For example, the processor (130) can calculate the mean and variance of the input values ​​and the mean and variance of the learning values.

[0081] In operation S230, the processor (130) can identify whether data drift has occurred in the artificial intelligence model (111) using the identified characteristics.

[0082] For example, the processor (130) can identify that data drift has occurred by comparing the average of input values ​​with the average of learning values, and if the difference between the averages is greater than a threshold value. In addition, the processor (130) can identify that data drift has occurred by comparing the variance of input values ​​with the variance of learning values, and if the difference between the variances is greater than a threshold value.

[0083] In this way, the processor (130) can identify that data drift has occurred if at least one of the average and variance of the input values ​​is different from at least one of the average and variance of the learning values.

[0084] In an embodiment, the process of comparing input values ​​with learning values ​​may be performed for each feature of the input values ​​of the artificial intelligence model. If the artificial intelligence model (111) has multiple features, the processor (130) may compare the input values ​​corresponding to each feature with the learning values ​​to identify whether data drift has occurred for each feature.

[0085] For example, it is assumed that the electronic device (100) is a refrigerator, and the artificial intelligence model (111) uses the rotation speed of the fan motor and the rotation speed of the compressor motor as input values ​​to predict the temperature of the storage room after a certain period of time. The processor (130) can identify whether data drift has occurred with respect to the rotation speed of the fan motor by comparing the rotation speed of the fan motor detected using the sensor (120) with the rotation speed of the fan motor used for learning. In addition, the processor (130) can identify whether data drift has occurred with respect to the rotation speed of the compressor motor by comparing the rotation speed of the compressor motor detected using the sensor (120) with the rotation speed of the compressor motor used for learning.

[0086] Meanwhile, although Fig. 2 describes identifying whether data drift has occurred by using the mean and variance of input values ​​and learning values, it is not limited thereto. For example, the processor (130) can identify whether data drift has occurred by comparing the distributions of input values ​​and learning values ​​using a statistical verification method (e.g., Kolmogorov-Smirnov test, Mann-Whitney U test, etc.).

[0087] In some embodiments, the evaluation of the predictive performance of the artificial intelligence model (111) may be performed in the electronic device (100).

[0088] For example, in the case of an artificial intelligence model (111) trained to identify whether a robot cleaner (100) is driving on a carpet, the evaluation of the predictive performance of the artificial intelligence model (111) cannot be performed in the electronic device (100) because the robot cleaner (100) cannot identify whether the robot cleaner (100) is driving on an actual cafe using other methods. However, in the case of an artificial intelligence model (111) trained to predict the temperature of a storage compartment after a certain period of time, the evaluation of the predictive performance of the artificial intelligence model (111) can be performed in the electronic device (100) because the refrigerator (100) can measure the actual temperature of the storage compartment after a certain period of time using a temperature sensor.

[0089] In this way, when the evaluation of the prediction performance of the artificial intelligence model (111) can be performed in the electronic device (100), the processor (130) identifies the prediction performance of the performance of the artificial intelligence model (111) using the actual value measured using the sensor (120) and the value predicted by the artificial intelligence model (111), and when it is identified that the prediction performance of the artificial intelligence model (111) has deteriorated, it is possible to identify whether data drift has occurred with respect to the artificial intelligence model.

[0090] For example, assume that the electronic device (100) is a refrigerator (100) and the artificial intelligence model (111) is a model trained to predict the temperature of the storage room after a certain period of time.

[0091] The processor (130) inputs the rotation speed of the fan motor and the rotation speed of the compressor motor detected by the sensor (120) into the artificial intelligence model (111), thereby obtaining information on the temperature of the storage room after a certain period of time from the artificial intelligence model (111). In addition, the processor (130) can measure the actual temperature of the storage room after a certain period of time using a temperature sensor for detecting the temperature of the storage room.

[0092] In addition, the processor (130) can identify the prediction performance of the artificial intelligence model using the measured temperature and the temperature predicted by the artificial intelligence model (111). For example, the processor (130) can obtain a value representing the prediction performance of the artificial intelligence model using a method such as RMSE, F1 score, etc., and can identify whether the prediction performance of the artificial intelligence model (111) has deteriorated based on the obtained value. For example, if the value obtained by RMSE is greater than a threshold value, the processor (130) can identify that the prediction performance of the artificial intelligence model (111) has deteriorated, and if the value obtained by RMSE is less than or equal to the threshold value, the processor (130) can identify that the prediction performance of the artificial intelligence model (111) has not deteriorated. In addition, if the processor (130) identifies that the prediction performance of the artificial intelligence model (111) has deteriorated, the processor (130) can identify whether data drift has occurred with respect to the artificial intelligence model (111).

[0093] In some embodiments, even if a component of an electronic device (100) fails, the output value of the component may differ from its normal state. Therefore, data drift due to component aging and component failure need to be distinguished.

[0094] For example, when a component ages, the output value of the component gradually changes over time, but when the component fails, the output value of the component may change more rapidly than before. Therefore, the processor (130) can acquire the output values ​​of the component using the sensor (120) over a certain period of time. Then, the processor (130) calculates a variance for the acquired values, and if the calculated variance is greater than a threshold value, the processor (130) identifies the component as failed, and if the calculated variance is less than the threshold value, the processor can identify whether data drift has occurred using the acquired values.

[0095] FIG. 3 is a flowchart illustrating an operation of an electronic device performing compensation for data drift according to an embodiment of the present disclosure.

[0096] In operations S310-Y and S320, if the processor (130) identifies that data drift has occurred, it can obtain a compensation value for compensating for the data drift.

[0097] In operation S330, the processor (130) can perform compensation for data drift based on the compensation value.

[0098] In one example, compensation for data drift may include compensation for input values ​​of an artificial intelligence model. For example, the processor (130) may obtain a first compensation value for compensating for the input values ​​of the artificial intelligence model, input the compensated input values ​​based on the first compensation value into the artificial intelligence model, and obtain the predicted values ​​of the artificial intelligence model from the artificial intelligence model.

[0099] In one example, compensation for data drift may include compensation for the predicted value of the artificial intelligence model. For example, the processor (130) may obtain a second compensation value for compensating for the predicted value of the artificial intelligence model, input the input value obtained using the sensor (120) into the artificial intelligence model to obtain the predicted value of the artificial intelligence model, and compensate for the predicted value of the artificial intelligence model based on the second compensation value.

[0100] In one embodiment, the processor (130) may include a compensation value based on the accumulated usage time of the electronic device (100). In this case, the compensation value may include a first compensation value for compensating an input value of the artificial intelligence model or a second compensation value for compensating a predicted value of the artificial intelligence model.

[0101] The memory (110) may store information regarding a plurality of compensation values ​​corresponding to a plurality of accumulated usage times. The processor (130) may obtain a compensation value corresponding to the accumulated usage time of the electronic device (100) from among the plurality of compensation values ​​corresponding to the plurality of accumulated usage times, and may perform compensation for data drift based on the compensation value.

[0102] For example, as the accumulated usage time of the electronic device (100) increases, the electronic device (100) may gradually age, and the difference between the output value of a component changed due to aging and the output value in a normal state may gradually increase. In this case, the change in the output value of a component due to aging may include an increase in the output value of the component compared to a normal state, or a decrease in the output value of the component compared to a normal state. Therefore, in the manufacturing stage of the electronic device (100), the change in the output value of each component according to the accumulated usage time of the electronic device (100) may be obtained through simulation, and a compensation value for each component may be obtained based on the change in the output value of each component and stored in the memory (110).

[0103] According to one example, a first compensation value for compensating for a changed output value of a component may be stored in the memory (110) so that the output value of the component changed due to aging becomes the same as the output value in a normal state. When the processor (130) determines that data drift has occurred, the processor (130) may identify a time range to which the accumulated usage time of the electronic device (100) belongs among a plurality of accumulated usage time ranges, and may obtain a first compensation value corresponding to the identified time range. In addition, the processor (130) may compensate for the output value of the component detected using the sensor (120) based on the first compensation value, and use the compensated output value as an input value of an artificial intelligence model.

[0104] For example, it is assumed that the electronic device (100) is a robot vacuum cleaner (100), and an artificial intelligence model (111) is trained to predict whether the robot vacuum cleaner (100) is driving on a carpet by using the acceleration value and the angular velocity value of the robot vacuum cleaner (100) as input values. In this case, as shown in FIGS. 4A and 4B, information on a compensation value for compensating the acceleration value and a compensation value for compensating the angular velocity value can be stored in the memory (110).

[0105] For example, it is assumed that data drift occurs for acceleration values ​​and the accumulated usage time of the robot cleaner (100) falls within the time range of t2 ≤ t < t3. The processor (130) can detect the acceleration value and angular velocity value of the robot cleaner (100) using the sensor (120). Then, the processor (130) can obtain a compensation value a2 for compensating for the acceleration value in which data drift occurs based on the accumulated usage time of the robot cleaner (100), and apply the compensation value a2 to the acceleration value detected using the sensor (120). For example, the processor (130) can multiply the compensation value a2 by the acceleration value. However, the present invention is not limited thereto, and one of the four arithmetic operations can be used. In addition, the processor (130) inputs the acceleration value to which the compensation value is applied and the angular velocity value detected using the sensor (120) into the artificial intelligence model (111), thereby identifying whether the robot cleaner (100) is driving on a carpet.

[0106] In this way, in the present disclosure, the input values ​​of the artificial intelligence model can be compensated for to compensate for data drift.

[0107] According to one example, a second compensation value for compensating for the predicted value of the artificial intelligence model may be stored in the memory (110) so that the predicted value of the artificial intelligence model using the output value of the component changed due to aging becomes the same as the predicted value of the artificial intelligence model using the output value of the normal state. When the processor (130) determines that data drift has occurred, the processor (130) may identify a time range to which the accumulated usage time of the electronic device (100) belongs among a plurality of accumulated usage time ranges, and may obtain a second compensation value corresponding to the identified time range. In addition, the processor (130) may use the output value of the component detected using the sensor (120) as an input value of the artificial intelligence model (111), obtain the predicted value of the artificial intelligence model (111) from the artificial intelligence model (111), and apply the second compensation value to the predicted value of the artificial intelligence model (111) to compensate for the predicted value of the artificial intelligence model (111).

[0108] For example, it is assumed that the electronic device (100) is a refrigerator (100), and the artificial intelligence model (111) is trained to predict the temperature of the storage compartment after a certain period of time using the rotation speed of the fan motor and the rotation speed of the compressor motor as input values. In this case, as shown in FIG. 5, information on a compensation value for compensating for the temperature of the storage compartment predicted by the artificial intelligence model (111) can be stored in the memory (110).

[0109] For example, it is assumed that data drift occurs for the rotation speed of the fan motor and the accumulated usage time of the refrigerator (100) falls within the time range of t2 ≤ t < t3. The processor (130) detects the rotation speed of the fan motor and the rotation speed of the compressor motor using the sensor (120), inputs the rotation speed of the fan motor and the rotation speed of the compressor motor into the artificial intelligence model (111), and obtains information on the temperature of the storage compartment after a certain period of time predicted by the artificial intelligence model (111) from the artificial intelligence model (111). In addition, the processor (130) may obtain a compensation value c2 for compensating for the temperature predicted by the artificial intelligence model (111) based on the accumulated usage time of the refrigerator (100), and apply the compensation value c2 to the predicted temperature. For example, the processor (130) may multiply the compensation value c2 by the predicted temperature. However, the present invention is not limited thereto, and one of the four arithmetic operations may be used. And, the processor (130) can determine that the compensated temperature is the temperature of the storage room after a certain period of time.

[0110] In this way, in the present disclosure, the predicted value of the artificial intelligence model can be compensated for data drift.

[0111] Meanwhile, while the aforementioned example described compensation values ​​as being stored in the memory (110) in the form of a lookup table, the present invention is not limited thereto. For example, a mathematical formula for calculating compensation values ​​for each component may be stored in the memory (110). The mathematical formula may have as variables the output value of the component and the accumulated usage time of the electronic device (100).

[0112] In one embodiment, the processor (130) may obtain a compensation value using the predicted value of the artificial intelligence model and the actual value detected using the sensor (120). In this case, the compensation value may include a first compensation value for compensating the input value of the artificial intelligence model or a second compensation value for compensating the predicted value of the artificial intelligence model.

[0113] In one example, the processor (130) can identify a first reward value using a plurality of candidate reward values.

[0114] For example, the processor (130) can obtain input values ​​of the artificial intelligence model (111). The input values ​​of the artificial intelligence model (111) can include data detected by the sensor (120). In this case, the processor (130) can collect input values ​​of the artificial intelligence model (111) using the sensor (120) for a certain period of time.

[0115] In addition, the processor (130) can obtain multiple input values ​​by applying each of the plurality of candidate reward values ​​to the input value of the artificial intelligence model (111). The candidate reward values ​​may be predefined values ​​or randomly selected values. For example, the processor (130) can multiply each of the plurality of candidate reward values ​​by the input value of the artificial intelligence model (111). However, the present invention is not limited thereto, and one of the four basic arithmetic operations may be used. In addition, the processor (130) can obtain multiple input values ​​by applying each of the plurality of candidate reward values ​​to the input value for each feature.

[0116] The processor (130) inputs each of a plurality of input values ​​into the artificial intelligence model (111), thereby obtaining a plurality of predicted values ​​of the artificial intelligence model (111), and can identify the predicted performance of the artificial intelligence model (111) corresponding to each of a plurality of candidate reward values ​​based on each of the plurality of predicted values ​​and the actual values ​​detected using the sensor (120). The predicted performance can be identified by a method such as, for example, RMSE, F1 score, etc.

[0117] In addition, the processor (130) may determine a reward value from among a plurality of candidate reward values ​​based on the prediction performance of the artificial intelligence model (111). For example, the processor (130) may determine a candidate reward value having the best prediction performance from among the prediction performances of the artificial intelligence model (111) corresponding to each of the plurality of candidate reward values ​​as the reward value.

[0118] For example, it is assumed that the electronic device (100) is a refrigerator (100), and the artificial intelligence model is a model trained to predict the temperature of the storage room after a certain period of time by using the rotation speed of the fan motor and the rotation speed of the compressor motor as input values.

[0119] In this case, if the processor (130) identifies that a data drift has occurred for the rotation speed of the fan motor, the processor (130) may obtain a plurality of candidate compensation values ​​for the rotation speed of the fan motor, and apply each of the plurality of candidate compensation values ​​to the rotation speed of the fan motor detected using the sensor (120), thereby obtaining a plurality of rotation speeds. For example, if the plurality of candidate compensation values ​​include a first candidate compensation value, a second candidate compensation value, and a third candidate compensation value, the processor (130) may obtain a rotation speed to which the first candidate compensation value is applied, a rotation speed to which the second candidate compensation value is applied, and a rotation speed to which the third candidate compensation value is applied.

[0120] The processor (130) inputs the rotation speed of the fan motor to which the first candidate compensation value is applied and the rotation speed of the compressor motor detected using the sensor (120) into the artificial intelligence model (111), and can obtain the temperature of the storage room after a certain period of time predicted by the artificial intelligence model (111) from the artificial intelligence model (111). In addition, the processor (130) can measure the actual temperature of the storage room after a certain period of time using the sensor (120), and compare the measured temperature with the temperature predicted by the artificial intelligence model (111) to identify the prediction performance of the artificial intelligence model (111).

[0121] In addition, the processor (130) inputs the rotation speed of the fan motor to which the second candidate compensation value is applied and the rotation speed of the compressor motor detected using the sensor (120) into the artificial intelligence model (111), so that the temperature of the storage room after a certain period of time predicted by the artificial intelligence model (111) can be obtained from the artificial intelligence model (111). In addition, the processor (130) measures the actual temperature of the storage room after a certain period of time using the sensor (120), and compares the measured temperature with the temperature predicted by the artificial intelligence model (111) to identify the prediction performance of the artificial intelligence model.

[0122] In addition, the processor (130) inputs the rotation speed of the fan motor to which the third candidate compensation value is applied and the rotation speed of the compressor motor detected using the sensor (120) into the artificial intelligence model (111), so that the temperature of the storage room after a certain period of time predicted by the artificial intelligence model can be obtained from the artificial intelligence model (111). In addition, the processor (130) measures the actual temperature of the storage room after a certain period of time using the sensor (120), and compares the measured temperature with the temperature predicted by the artificial intelligence model to identify the prediction performance of the artificial intelligence model (111).

[0123] And, the processor (130) can determine the candidate compensation value with the best prediction performance among the first to third candidate compensation values ​​as a compensation value for compensating the rotation speed of the fan motor. For example, as shown in FIG. 6, when the first candidate compensation value is applied, the value calculated by RMSE is s1, when the second candidate compensation value is applied, the value calculated by RMSE is s2, and when the third candidate compensation value is applied, the value calculated by RMSE is s3. In this case, s2 <s1<s3이다. 이 경우, RMSE에 의해 산출된 값들 중에서 가장 작은 값일 때, 인공지능 모델(111)의 예측 성능이 가장 높은 것으로 볼 수 있다. 따라서, 프로세서(130)는 RMSE에 의해 산출된 값들 중 제2 후보 보상 값을 보상 값으로 결정할 수 있다.

[0124] Thereafter, the processor (130) compensates for the rotation speed of the fan motor detected using the sensor (120) using the compensation value, and inputs the compensated rotation speed of the fan motor and the rotation speed of the compressor motor detected using the sensor (120) into the artificial intelligence model (111), so that information on the temperature of the storage room after a certain period of time predicted by the artificial intelligence model (111) can be obtained from the artificial intelligence model (111).

[0125] According to one example, the processor (130) can identify a second compensation value for compensating the predicted value of the artificial intelligence model (111) so that the predicted value of the artificial intelligence model (111) becomes the same as the actual value detected using the sensor (120).

[0126] For example, the processor (130) can input the input value of the artificial intelligence model (111) obtained using the sensor (120) into the artificial intelligence model, thereby obtaining the predicted value of the artificial intelligence model (111) from the artificial intelligence model (111).

[0127] In addition, the processor (130) can obtain a compensation value to be applied to the predicted value of the artificial intelligence model (111) so that the predicted value of the artificial intelligence model (111) becomes the same as the actual value obtained using the sensor (120). Thereafter, the processor (130) inputs the input value of the artificial intelligence model (111) obtained using the sensor (120) into the artificial intelligence model (111), obtains the predicted value of the artificial intelligence model (111), and applies the compensation value to the predicted value to compensate for the predicted value.

[0128] For example, it is assumed that the electronic device (100) is a refrigerator (100), and the artificial intelligence model is a model trained to predict the temperature of the storage compartment after a certain period of time by using the rotation speed of the fan motor and the rotation speed of the compressor motor as input values. In this case, the processor (130) inputs the rotation speed of the fan motor and the rotation speed of the compressor motor detected by the sensor (120) into the artificial intelligence model (111), and can obtain information about the temperature of the storage compartment after a certain period of time predicted by the artificial intelligence model (111) from the artificial intelligence model (111). In addition, the processor (130) can measure the actual temperature of the storage compartment after a certain period of time by using the sensor (120).

[0129] Thereafter, the processor (130) can identify a compensation value for compensating the predicted temperature so that the temperature predicted by the artificial intelligence model (111) becomes the same as the measured actual temperature. Thereafter, the processor (130) can input the rotation speed of the fan motor and the rotation speed of the compressor motor detected by the sensor (120) into the artificial intelligence model (111) to obtain the predicted temperature from the artificial intelligence model (111). Then, the processor (130) can apply the compensation value to the predicted temperature to compensate for the predicted temperature, and identify the compensated temperature as the temperature of the storage room after a certain period of time.

[0130] In one embodiment, the processor (130) may transform the input values ​​so that the mean and variance of the input values ​​of the artificial intelligence model (111) become the same as the mean and variance of the learning values ​​used to train the artificial intelligence model (111), and input the transformed input values ​​into the artificial intelligence model (111) to obtain the predicted values ​​of the artificial intelligence model (111). As described above, the data drift may occur due to the difference in statistical characteristics between the learning values ​​of the artificial intelligence model and the input values ​​of the artificial intelligence model. Therefore, the processor (130) may transform the distribution of the input values ​​so that the mean and variance of the input values ​​become the mean and variance of the learning values, and input the transformed input values ​​into the artificial intelligence model (111) to obtain the predicted values ​​of the artificial intelligence model (111) from the artificial intelligence model (111).

[0131] In one embodiment, the processor (130) may change a threshold value compared to the predicted value of the artificial intelligence model (111) to compensate for data drift.

[0132] For example, the artificial intelligence model (111) may use the output value of a component detected by a sensor (120) as an input value. At this time, the prediction performance of the artificial intelligence model may deteriorate due to changes in the output value of the component caused by aging. Therefore, the processor (130) may change the threshold value compared with the predicted value of the artificial intelligence model (111) to prevent incorrect predictions by the artificial intelligence model.

[0133] For example, it is assumed that the electronic device (100) is a robot cleaner (100), and the artificial intelligence model (111) is a model trained to predict whether the robot cleaner (100) is driving on a carpet by using the acceleration value and the angular velocity value of the robot cleaner (100) as input values. The processor (130) can identify whether the robot cleaner (100) is driving on a carpet by comparing the probability value obtained from the artificial intelligence model (111) with a threshold value. For example, if the probability value obtained from the artificial intelligence model (111) is greater than or equal to the threshold value, the processor (130) can identify that the robot cleaner (100) is driving on a carpet. The threshold value may be, for example, th1. Meanwhile, if the processor (130) determines that data drift has occurred, the threshold value may be adjusted from th1 to th2. At this time, th1 <th2일 수 있다. 그리고, 프로세서(130)는 인공지능 모델(111)로부터 획득한 확률 값과 임계 값을 비교하여, 확률 값이 임계 값 이상이면, 로봇 청소기(100)가 카펫 위를 주행하는 것으로 식별할 수 있다.

[0134] In this way, in the present disclosure, when data drift occurs, the input value of the artificial intelligence model or the predicted value of the artificial intelligence model is compensated, thereby preventing a decline in the performance of the artificial intelligence model without updating the artificial intelligence model.

[0135] FIG. 7 is a flowchart illustrating a data drift compensation method of an electronic device in which an artificial intelligence model is stored according to an embodiment of the present disclosure.

[0136] In operation S710, if it is determined that data drift has occurred for the artificial intelligence model based on the input values ​​of the artificial intelligence model acquired using the sensor, a compensation value for compensating for the data drift is acquired.

[0137] In operation S720, compensation for data drift is performed based on the compensation value.

[0138] For example, in operation S710, a first compensation value for compensating an input value of an artificial intelligence model can be obtained. Furthermore, in operation S720, an input value compensated based on the first compensation value can be input into the artificial intelligence model, thereby obtaining a predicted value of the artificial intelligence model from the artificial intelligence model.

[0139] For example, in operation S710, a second compensation value may be obtained to compensate for the predicted value of the artificial intelligence model. Furthermore, in operation S720, the predicted value of the artificial intelligence model obtained by inputting an input value into the artificial intelligence model may be obtained from the artificial intelligence model, and the predicted value may be compensated based on the second compensation value.

[0140] Additionally, the electronic device may store information regarding a plurality of compensation values ​​corresponding to a plurality of accumulated usage times. In operation S710, a compensation value corresponding to the accumulated usage time of the electronic device may be acquired from among the plurality of compensation values ​​corresponding to the plurality of accumulated usage times.

[0141] Additionally, in operation S710, a compensation value can be obtained using the predicted value of the artificial intelligence model and the actual value detected using the sensor.

[0142] In addition, in operation S710, a plurality of candidate reward values ​​are applied to input values ​​of an artificial intelligence model to obtain a plurality of input values, a plurality of input values ​​are input to an artificial intelligence model to obtain a plurality of predicted values ​​of the artificial intelligence model, and a prediction performance of an artificial intelligence model corresponding to each of the plurality of candidate reward values ​​is identified based on each of the plurality of predicted values ​​and an actual value detected using a sensor, and a reward value is determined from among the plurality of candidate reward values ​​based on the identified predicted performance.

[0143] Additionally, the method according to the present disclosure may further include a step of identifying whether data drift has occurred based on the learning value of the artificial intelligence model and the input value of the artificial intelligence model obtained using a sensor.

[0144] Meanwhile, various embodiments of the present disclosure may be implemented in a computer-readable recording medium or similar device using software, hardware, or a combination thereof. In some cases, the embodiments described herein may be implemented by the processor itself. In a software implementation, embodiments, such as the procedures and functions described herein, may be implemented as separate software modules. Each of the software modules may perform one or more functions and operations described herein.

[0145] Meanwhile, computer instructions for performing processing operations of an electronic device according to various embodiments of the present disclosure described above may be stored in a non-transitory computer-readable medium. When the computer instructions stored in such a non-transitory computer-readable medium are executed by a processor of a specific device, the computer instructions cause the specific device to perform processing operations in a robot cleaner (100) according to various embodiments described above.

[0146] A non-transitory computer-readable medium refers to a medium that permanently stores data and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specific examples of non-transitory computer-readable media include CDs, DVDs, hard disks, Blu-ray discs, USBs, memory cards, and ROMs.

[0147] Although the preferred embodiments of the present disclosure have been illustrated and described above, the present disclosure is not limited to the specific embodiments described above, and various modifications may be made by a person having ordinary skill in the art to which the present disclosure pertains without departing from the gist of the present disclosure as claimed in the claims, and such modifications should not be understood individually from the technical idea or prospect of the present disclosure.

Claims

1. In electronic devices, sensor; Memory where artificial intelligence models are stored; and If it is determined that data drift has occurred for the artificial intelligence model based on the input values of the artificial intelligence model acquired using the sensor, a compensation value for compensating for the data drift is acquired, An electronic device comprising: one or more processors for performing compensation for the data drift based on the compensation value; 2. In paragraph 1, One or more of the above processors, An electronic device that obtains a first compensation value for compensating an input value of the artificial intelligence model, and inputs the compensated input value based on the first compensation value into the artificial intelligence model to obtain a predicted value of the artificial intelligence model from the artificial intelligence model.

3. In paragraph 1, One or more of the above processors, An electronic device that obtains a second compensation value for compensating for a predicted value of the artificial intelligence model, obtains a predicted value of the artificial intelligence model obtained by inputting the input value into the artificial intelligence model, and compensates for the predicted value based on the second compensation value.

4. In paragraph 1, The above memory stores information about multiple reward values corresponding to multiple accumulated usage times, One or more of the above processors, An electronic device that obtains a compensation value corresponding to the accumulated usage time of the electronic device from among a plurality of compensation values corresponding to the plurality of accumulated usage times.

5. In paragraph 1, One or more of the above processors, An electronic device that obtains the compensation value using the predicted value of the artificial intelligence model and the actual value detected using the sensor.

6. In paragraph 5, One or more of the above processors, Applying each of the multiple candidate reward values to the input values of the artificial intelligence model to obtain multiple input values, By inputting each of the plurality of input values into the artificial intelligence model, a plurality of predicted values of the artificial intelligence model are obtained, Identifying the prediction performance of the artificial intelligence model corresponding to each of the plurality of candidate reward values based on each of the plurality of predicted values and the actual value detected using the sensor, An electronic device that determines the reward value from among the plurality of candidate reward values based on the identified prediction performance.

7. In paragraph 1, One or more of the above processors, An electronic device that identifies whether data drift has occurred based on the learning value of the artificial intelligence model and the input value of the artificial intelligence model obtained using the sensor.

8. In a method for compensating for data drift of an electronic device in which an artificial intelligence model is stored, A step of obtaining a compensation value for compensating for the data drift when it is identified that data drift has occurred for the artificial intelligence model based on the input value of the artificial intelligence model acquired using a sensor; and A compensation method comprising: a step of performing compensation for the data drift based on the compensation value; 9. In paragraph 8, The above acquisition steps are: A step of obtaining a first compensation value for compensating the input value of the artificial intelligence model; The steps performed above are: A compensation method comprising: a step of inputting a compensated input value based on the first compensation value into the artificial intelligence model and obtaining a predicted value of the artificial intelligence model from the artificial intelligence model.

10. In paragraph 8, The above acquisition steps are: A step of obtaining a second compensation value for compensating the predicted value of the artificial intelligence model; The steps performed above are: A compensation method comprising: a step of obtaining a predicted value of the artificial intelligence model obtained by inputting the input value into the artificial intelligence model, and compensating the predicted value based on the second compensation value.

11. In paragraph 8, The electronic device stores information about a plurality of compensation values corresponding to a plurality of accumulated usage times, The above acquisition steps are: A compensation method for obtaining a compensation value corresponding to the accumulated usage time of the electronic device from among a plurality of compensation values corresponding to the plurality of accumulated usage times.

12. In paragraph 8, The above acquisition steps are: A compensation method for obtaining the compensation value using the predicted value of the artificial intelligence model and the actual value detected using the sensor.

13. In paragraph 12, The above acquisition steps are: A step of obtaining multiple input values by applying each of the multiple candidate reward values to the input values of the artificial intelligence model; A step of inputting each of the plurality of input values into the artificial intelligence model to obtain a plurality of predicted values of the artificial intelligence model; A step of identifying the prediction performance of the artificial intelligence model corresponding to each of the plurality of candidate reward values based on each of the plurality of predicted values and the actual value detected using the sensor; and A compensation method comprising: a step of determining the compensation value from among the plurality of candidate compensation values based on the identified prediction performance.

14. In paragraph 8, A compensation method further comprising a step of identifying whether the data drift has occurred based on the learning value of the artificial intelligence model and the input value of the artificial intelligence model obtained using the sensor.

Images