Information processing device and non-transitory computer-readable medium

Abstract

An information processing device and a non-transitory computer-readable medium. An information processing device includes a controller that causes learning data learned by an artificial intelligence to be recorded in a recording unit in a manner that allows for distinguishing between influential learning data that has an impact on the performance of the artificial intelligence and non-influential learning data that has no impact on the performance of the artificial intelligence.

Description

Technical Field

[0001] The present disclosure relates to an information processing device and a non-transitory computer-readable medium. Background Art

[0002] Typically, learning data is learned by artificial intelligence.

[0003] Japanese Unexamined Patent Application Publication No. 2010-223824 describes an apparatus for preventing repeated learning of an erroneous new path. Summary of the Invention

[0004] Therefore, the purpose of this disclosure is to find out whether learning data has an impact on the performance of artificial intelligence.

[0005] According to the first embodiment of the present disclosure, an information processing device is provided, which includes a controller, which enables learning data learned by artificial intelligence to be recorded in a recording unit in a manner that can distinguish between influential learning data that has an impact on the performance of artificial intelligence and non-influence learning data that has no impact on the performance of artificial intelligence.

[0006] According to the second aspect of the present disclosure, the information processing device is configured so that the controller causes the learning data to be recorded in the recording unit in a manner that enables improvement data that improves the performance of artificial intelligence to be distinguished from reduction data that reduces the performance of artificial intelligence to be distinguished.

[0007] According to a third aspect of the present disclosure, the information processing device is configured so that the controller causes the learning data to be recorded in the recording unit in such a manner that learning data learned by different artificial intelligences can be distinguished.

[0008] According to a fourth aspect of the present disclosure, the information processing device is configured so that the controller causes the learning data to be recorded in the recording unit in such a manner that learning data learned by different artificial intelligences can be distinguished.

[0009] According to the fifth aspect of the present disclosure, the information processing device is configured so that the controller causes the learning data to be recorded in the recording unit in a manner such that the influential learning data is associated with an affected function of the artificial intelligence that is affected by the influential learning data.

[0010] According to a sixth aspect of the present disclosure, the information processing apparatus is configured so that the controller causes the learning data to be recorded in the recording unit in association with a learning history of the artificial intelligence that has learned the learning data.

[0011] According to a seventh aspect of the present disclosure, the information processing apparatus is configured so that the controller causes the learning data to be recorded in the recording unit in such a manner that influential combinations of learning data sets can be distinguished from non-influential combinations of learning data sets.

[0012] According to an eighth aspect of the present disclosure, the information processing device is configured so that, depending on a learning purpose of another artificial intelligence, the controller causes the other artificial intelligence to learn the influential learning data or the non-influence learning data.

[0013] According to a ninth aspect of the present disclosure, the information processing device is configured so that, depending on a learning purpose of another artificial intelligence, the controller causes the another artificial intelligence to learn improved learning data or not improved learning data.

[0014] According to a tenth aspect of the present disclosure, the information processing device is configured so that the controller causes another artificial intelligence to learn influential learning data that has an influence on performance of an artificial intelligence whose function corresponds to the function of the another artificial intelligence.

[0015] According to the eleventh embodiment of the present disclosure, an information processing device is provided, which includes a controller, which enables learning data learned by artificial intelligence to be recorded in a recording unit in a manner that can distinguish between improved learning data that improves the performance of artificial intelligence and non-improvement learning data that does not improve the performance of artificial intelligence.

[0016] According to the twelfth aspect of the present disclosure, the information processing device is configured so that the controller causes the learning data to be recorded in the recording unit in a manner such that the improved learning data is associated with the improved function of the artificial intelligence improved by the improved learning data.

[0017] According to a thirteenth aspect of the present disclosure, the information processing device is configured so that the controller causes another artificial intelligence to learn improvement learning data that improves the performance of an artificial intelligence whose function corresponds to the function of the another artificial intelligence.

[0018] According to the fourteenth embodiment of the present disclosure, a non-temporary computer-readable medium is provided, which stores a program for causing a computer to perform an image processing process, wherein the process includes causing learning data learned by artificial intelligence to be recorded in a recording unit in a manner that can distinguish influential learning data that has an impact on the performance of artificial intelligence from non-influence learning data that has no impact on the performance of artificial intelligence.

[0019] According to the fifteenth embodiment of the present disclosure, a non-temporary computer-readable medium is provided, which stores a program for causing a computer to perform an image processing process, wherein the process includes causing learning data learned by artificial intelligence to be recorded in a recording unit in a manner that enables improved learning data that improves the performance of artificial intelligence to be distinguished from non-improved learning data that does not improve the performance of artificial intelligence to be distinguished.

[0020] According to the first, third, and fourteenth aspects of the present disclosure, it is possible to clearly determine whether learning data has an impact on the performance of artificial intelligence.

[0021] According to the second, fourth, eleventh, and fifteenth aspects of the present disclosure, it is possible to clearly determine whether the learning data improves the performance of artificial intelligence.

[0022] According to the fifth aspect of the present disclosure, it is possible to clearly determine whether learning data has an impact on the performance of artificial intelligence for each function of artificial intelligence.

[0023] According to the sixth aspect of the present disclosure, it is possible to manage the learning history of artificial intelligence.

[0024] According to the seventh solution of the present disclosure, it is possible to clearly determine whether the combination of different learning data sets has an impact on the performance of artificial intelligence.

[0025] According to the eighth and ninth aspects of the present disclosure, another artificial intelligence can be caused to learn learning data according to the purpose of learning.

[0026] According to the tenth aspect of the present disclosure, it is possible to obtain a learning effect obtained by learning the influential learning data through another artificial intelligence.

[0027] According to the twelfth aspect of the present disclosure, it is possible to clarify whether the learning data improves the performance of the artificial intelligence for each function of the artificial intelligence.

[0028] According to the thirteenth aspect of the present disclosure, the performance of another artificial intelligence can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Exemplary embodiments of the present disclosure will be described in detail based on the following drawings, in which:

[0030] Figure 1 is a block diagram illustrating a configuration of an information processing apparatus according to a first exemplary embodiment;

[0031] Figure 2 The learning data according to the first exemplary embodiment is illustrated;

[0032] Figure 3 Content data according to the first exemplary embodiment is illustrated;

[0033] Figure 4 is a block diagram illustrating a configuration of an information processing apparatus according to a second exemplary embodiment;

[0034] Figure 5 A learning data set according to a second exemplary embodiment is illustrated;

[0035] Figure 6 A learning data set according to a second exemplary embodiment is illustrated;

[0036] Figure 7 A learning data set according to a second exemplary embodiment is illustrated;

[0037] Figure 8 A database of determination results is illustrated;

[0038] Figure 9 A learning data set according to a first modification of the second exemplary embodiment is illustrated;

[0039] Figure 10 A learning data set according to a first modification of the second exemplary embodiment is illustrated;

[0040] Figure 11 A learning data set according to a first modification of the second exemplary embodiment is illustrated;

[0041] Figure 12 A database of determination results is illustrated;

[0042] Figure 13 illustrates a learning data set according to a second modification of the second exemplary embodiment; and

[0043] Figure 14 A learning data set according to a second modification of the second exemplary embodiment is illustrated. DETAILED DESCRIPTION

[0044] First Exemplary Embodiment

[0045] Will refer to Figure 1 An information processing apparatus according to a first exemplary embodiment of the present disclosure will be described. Figure 1 An example of the information processing apparatus according to the first exemplary embodiment is illustrated.

[0046] The information processing apparatus 10 according to the first exemplary embodiment is configured to receive learning data and control artificial intelligence to learn according to information attached to the learning data, the information indicating whether artificial intelligence (ie, AI) is permitted to learn the learning data.

[0047] The information processing device 10 is, for example, a personal computer (hereinafter referred to as a "PC"), a tablet PC, a smartphone, a mobile phone, or any other type of device (for example, a multifunction printer having functions such as a printing function). Needless to say, the information processing device 10 may be a device other than these devices.

[0048] The algorithm used in artificial intelligence is not particularly limited and can be any algorithm. For example, the algorithm can be machine learning. Machine learning can be supervised learning, unsupervised learning or reinforcement learning. Specifically, deep learning (for example, multilayer perceptron, convolutional neural network, recursive neural network, autoencoder, restricted Boltzmann machine), perceptron, back propagation, associator, support vector machine, decision tree, k nearest neighbor algorithm, linear regression, self-organizing map, Boltzmann machine, principal component analysis, cluster analysis, Q learning, etc. can be used. Genetic algorithm, hill climbing method, etc. can be used as algorithms other than machine learning. Needless to say, algorithms other than these algorithms can be used.

[0049] Learning data is data used for learning artificial intelligence. For supervised learning, learning data may include correct judgments (i.e., answers). For unsupervised learning, learning data may not necessarily include correct judgments as learning data. For example, learning data may be document data (e.g., text data), image data (e.g., still image data or moving image data), music data, audio data, or a combination thereof, and its type, data format, and content are not particularly limited.

[0050] Attribute information is attached to the learning data. The attribute information attached to the learning data is information indicating whether the artificial intelligence is permitted to learn the learning data. The state in which the attribute information is attached to the learning data means that the attribute information accompanies the learning data as accompanying information, or the attribute information is embedded in the learning data itself. The attribute information can be added to the learning data as accompanying information, or can be associated with the learning data as data separate from the learning data. For example, the state in which the attribute information is embedded in the learning data itself is a state in which the attribute information is set in an image represented by image data or a state in which the attribute information is set in a document represented by document data.

[0051] Artificial intelligence may be installed in the information processing device 10, or may be installed in a device other than the information processing device 10 (for example, a server or a PC). That is, a program for realizing artificial intelligence may be stored in the information processing device 10, or may be stored in a device other than the information processing device 10.

[0052] The configuration of the information processing apparatus 10 is described below in detail.

[0053] The communication unit 12 is a communication interface and has a function of sending information to another device and a function of receiving information from another device. The communication unit 12 may have a wireless communication function or a wired communication function. The communication unit 12 may communicate with another device via a communication path using wireless communication or wired communication. For example, the communication path is a network such as a local area network (LAN) or the Internet. The communication unit 12 may communicate with another device without a communication path by using, for example, short-range wireless communication. Short-range wireless communication is, for example, Bluetooth (registered trademark), radio frequency identifier (RFID) or NFC.

[0054] For example, when learning data is transmitted from a device other than the information processing device 10 to the information processing device 10, the communication unit 12 receives the learning data. When attribute information is attached to the learning data, the communication unit 12 also receives the attribute information. In addition, the communication unit 12 can transmit the learning data to another device.

[0055] The UI unit 14 is a user interface. Examples include a display and an operation unit. The display is a display device such as a liquid crystal display. The operation unit is an input unit such as a keyboard, input keys, or an operation panel. A touch panel that functions as both a display and an operation unit is also an example of the UI unit 14.

[0056] The storage unit 16 is one or more storage areas in which various types of information are stored. Each storage area is composed of, for example, one or more storage devices (e.g., physical drives such as hard disk drives or memories) provided in the information processing device 10. Learning data can be stored in the storage unit 16.

[0057] The receiving unit 18 is configured to receive learning data. For example, in the case where the communication unit 12 receives learning data sent from a device other than the information processing device 10, the receiving unit 18 receives the learning data received by the communication unit 12. In the case where the communication unit 12 receives attribute information attached to the learning data, the receiving unit 18 also receives the attribute information. In addition, in the case where the learning data is supplied to the information processing device 10 through the UI unit 14, or in the case where the learning data is supplied to the information processing device 10 by using a storage device (for example, a hard disk drive, a USB memory, a CD, a DVD, or any one of other portable storage media), the receiving unit 18 receives the supplied learning data. In the case where attribute information is attached to the supplied learning data, the receiving unit 18 also receives the attribute information.

[0058] The changing unit 20 is configured to change the learning process of artificial intelligence in accordance with the attribute information attached to the learning data.

[0059] If the attribute information attached to the learning data indicates that the artificial intelligence is permitted to learn the learning data, the change unit 20 permits the artificial intelligence to learn the learning data. For example, if the user instructs the artificial intelligence to learn the learning data by operating the UI unit 14, the controller 28 causes the artificial intelligence to learn the learning data. The user can specify a specific artificial intelligence to learn the learning data and cause the artificial intelligence to learn the learning data.

[0060] If the attribute information attached to the learning data indicates that the artificial intelligence is not permitted to learn the learning data, the change unit 20 prohibits the artificial intelligence from learning the learning data. That is, the change unit 20 does not permit the artificial intelligence to learn the learning data. For example, even if the user instructs the artificial intelligence to learn the learning data by operating the UI unit 14, the controller 28 does not allow the artificial intelligence to learn the learning data.

[0061] The attribute information attached to the learning data may indicate that only a specific portion of the learning data is permitted to be learned, but that the remaining portion of the learning data is prohibited from being learned. Alternatively, the attribute information attached to the learning data may indicate that all learning data is permitted or prohibited to be learned.

[0062] When the attribute information attached to the learning data indicates that the artificial intelligence is not permitted to learn the learning data, the change unit 20 may permit the artificial intelligence to learn the learning data if the purpose of the artificial intelligence learning the learning data satisfies a specific condition. For example, when the purpose of learning is a highly public purpose such as a medical purpose or an educational purpose, the change unit 20 may permit the artificial intelligence to learn the learning data.

[0063] The output unit 22 is configured to output a warning when an attempt is made to change the learning data without permission, if the attribute information of the learning data indicates that the artificial intelligence is not permitted to learn the learning data. An attempt to change the learning data without permission refers to an attempt by a user who does not have permission to change the learning data to change the learning data. The user can be granted permission to change any learning data set or a specific learning data set.

[0064] For example, the learning data may be associated with user identification information (e.g., user name, user ID, password) for identifying a user with authority to change the learning data. In the case where the user identification information of the user who is attempting to change the learning data is associated with the learning data, the controller 28 permits the user to change the learning data. Specifically, in the case where the user performs an operation for changing the learning data on the UI unit 14, the controller 28 accepts the operation and changes the learning data in accordance with the user's operation. In the case where the user identification information of the user who is attempting to change the learning data is not associated with the learning data, the controller 28 prohibits the user from changing the learning data. Specifically, in the case where the user performs an operation for changing the learning data on the UI unit 14, the controller 28 rejects the operation and does not change the learning data. Note that the operation for changing the learning data may be performed at a device other than the information processing device 10 (e.g., a PC).

[0065] When a user logs into an information processing device or an equivalent device, or when a user attempts to change learning data, the user's user identification information may be collected. For example, when a user specifies learning data by operating the UI unit 14, the controller 28 causes an input screen for inputting the user's user identification information to be displayed on the display of the UI unit 14. For another example, when a user logs into the information processing device 10, the controller 28 causes an input screen for inputting the user's user identification information to be displayed on the display of the UI unit 14. The controller 28 permits the user to change the learning data.

[0066] The output unit 22 can output the warning in various forms. For example, the output unit 22 can cause information indicating the warning to be displayed on the display of the UI unit 14, can send the information indicating the warning to a device (e.g., a PC) that has sent the learning data to the information processing device 10, can cause a speaker to generate a warning sound, or can send the information indicating the warning to a pre-registered device (e.g., a PC used by an administrator, etc.).

[0067] The deletion unit 24 is configured to delete the learning data when an attempt is made to change the learning data without permission when the attribute information attached to the learning data indicates that the artificial intelligence is not allowed to learn the learning data. As described above, an attempt to change the learning data without permission means that a user who does not have the authority to change the learning data attempts to change the learning data.

[0068] For example, in the case where an attempt is made to change the learning data stored in the storage unit 16 without permission, the deletion unit 24 deletes the learning data from the storage unit 16. In the case where an attempt is made to change the learning data stored in a device other than the information processing device 10 (e.g., a PC or a server) without permission, the deletion unit 24 may delete the learning data from the other device.

[0069] The deleting unit 24 may delete the learning data received by the receiving unit 18 that is not permitted for artificial intelligence learning.

[0070] The attaching unit 26 is configured to attach attribute information to content data. The content data is, for example, document data (e.g., text data), image data (e.g., still image data or moving image data), music data, audio data, or a combination thereof, and its type, data format, and content are not particularly limited. The content data to which attribute information is attached is an example of learning data to which attribute information is attached.

[0071] The attaching unit 26 may not necessarily be provided in the information processing device 10, but may be provided in a device different from the device having the changing unit 20. In this case, the processing of the attaching unit 26 is performed by the device provided with the attaching unit 26.

[0072] The controller 28 is configured to control the operation of each unit of the information processing device 10 .

[0073] Below, refer to Figure 2 The processing of the information processing device 10 will be described. Figure 2 An example of learning data according to the first exemplary embodiment is illustrated.

[0074] Attribute information 32 is attached to learning data 30. Attribute information 32 is information indicating that artificial intelligence is permitted to learn learning data 30. Attribute information 36 is attached to learning data 34. Attribute information 36 is information indicating that artificial intelligence is not permitted to learn learning data 34.

[0075] In the learning data 30 will be artificial intelligence ( Figure 2 In a case where the artificial intelligence (“AI” in the example above) learns the learning data 30 (for example, in a case where the user gives an instruction to make the artificial intelligence learn the learning data 30), the changing unit 20 permits the artificial intelligence to learn the learning data 30 or prohibits the artificial intelligence from learning the learning data 30 based on the attribute information 32 attached to the learning data 30. Since the attribute information 32 is information indicating that the artificial intelligence is permitted to learn the learning data 30, the changing unit 20 permits the artificial intelligence to learn the learning data 30.

[0076] Similarly, in the case where learning data 34 is to be learned by artificial intelligence, the changing unit 20 permits the artificial intelligence to learn the learning data 34 or prohibits the artificial intelligence from learning the learning data 34 based on the attribute information 36 attached to the learning data 34. Since the attribute information 36 is information indicating that the artificial intelligence is not permitted to learn the learning data 34, the changing unit 20 prohibits the artificial intelligence from learning the learning data 34. This makes it possible to prevent learning data that is not intended to be learned by artificial intelligence from being learned by artificial intelligence.

[0077] Below, refer to Figure 3 Describe the process for creating learning data. Figure 3 An example of content data according to the first exemplary embodiment is illustrated.

[0078] Adding unit 26 adds attribute information to content data 38. Content data 38 to which attribute information is added is an example of learning data to which attribute information is added. Attribute information 40 indicates that AI learning is not permitted. If attribute information 40 is added to content data 38, AI learning of content data 38 is prohibited. Attribute information 42 indicates that AI learning is permitted. If attribute information 40 is added to content data 38, AI learning of content data 38 is permitted.

[0079] For example, by operating the UI unit 14, the user can specify the content data 38 to which the attribute information is to be attached, and can specify whether to permit the artificial intelligence learning content data 38. In the case where the user permits the artificial intelligence learning content data 38, the attachment unit 26 attaches the attribute information 42 to the content data 38 specified by the user. In the case where the user does not permit the artificial intelligence learning content data 38, the attachment unit 26 attaches the attribute information 40 to the content data 38 specified by the user. Needless to say, the attribute information can be attached to the learning data by a device other than the information processing device 10 (for example, a PC or a server).

[0080] Whether to permit artificial intelligence to learn content data 38 can be specified by the user who created content data 38, the user who provided content data 38, the user who purchased content data 38, the user who has permission to use content data 38, the administrator who manages content data 38, or a user other than these users.

[0081] For example, when a user creates content data 38, the appending unit 26 appends attribute information to the content data 38. The appending unit 26 may append attribute information to the content data 38 in accordance with an instruction from the user to append attribute information to the content data 38, or may append attribute information to the content data 38 when the content data 38 is completed without an instruction from the user.

[0082] Attachment unit 26 may attach attribute information to content data 38 regardless of the user's instructions. For example, if content data 38 includes an incorrect answer, attachment unit 26 may attach attribute information 40 indicating that artificial intelligence is not permitted to learn content data 38 to content data 38. For example, a case where content data 38 includes an incorrect answer is a case where content data 38 used as learning data for translation includes incorrect translation data, a case where content data 38 used as learning data for image recognition includes information indicating an object different from the object represented by content data 38, or a case where content data 38 used as learning data for character recognition includes information indicating a character different from the character represented by content data 38.

[0083] Attachment unit 26 determines whether content data 38 includes an incorrect answer by analyzing content data 38. If content data 38 includes an incorrect answer, attachment unit 26 attaches attribute information 40 indicating that artificial intelligence learning is not permitted to content data 38. Attachment unit 26 may attach attribute information 40 indicating that artificial intelligence learning is not permitted to content data 38 if the proportion of incorrect answers or the number of occurrences of incorrect answers is equal to or greater than a predetermined threshold.

[0084] If the content data 38 does not include an incorrect answer, the appending unit 26 appends attribute information 42 indicating that artificial intelligence learning is permitted to the content data 38. If the proportion of incorrect answers or the number of occurrences of incorrect answers is less than a predetermined threshold, the appending unit 26 may append attribute information 42 to the content data 38.

[0085] In the case where the content data 38 includes information about security, the appending unit 26 may append attribute information 40 indicating that artificial intelligence learning is not permitted to the content data 38. In the case where the content data 38 does not include information about security, the appending unit 26 may append attribute information 42 indicating that artificial intelligence learning is permitted to the content data 38. Examples of information about security include personal information, information about trade secrets, and information set as non-public information by a public institution (e.g., an administrative agency). Examples of personal information include information indicating a specific individual's name, date of birth, user ID, password, etc. The appending unit 26 determines whether the content data 38 includes information about security by analyzing the content data 38. If the content data 38 includes information about security, the attribute information 40 indicating that artificial intelligence learning is not permitted is appended to the content data 38, thereby preventing the information about security from being leaked due to artificial intelligence learning of the content data 38.

[0086] The change unit 20 can be configured to change the learning process of the artificial intelligence according to attribute information indicating a specific time period attached to the learning data. The attribute information attached to the learning data may include permitted time period information indicating a permitted time period during which the artificial intelligence is permitted to learn the learning data. That is, the attribute information can limit the time period during which the learning data to which the attribute information is attached is permitted. The artificial intelligence is prohibited from learning the learning data at any time outside the permitted time period. The time period during which the artificial intelligence is prohibited from learning the learning data can be referred to as a prohibited time period. For example, the permitted time period can be determined by date, time zone, time, etc. If the time point (e.g., date or time) at which the artificial intelligence is about to learn the learning data falls within the permitted time period, the change unit 20 permits the artificial intelligence to learn the learning data. If the time point at which the artificial intelligence is about to learn the learning data does not fall within the permitted time period, the change unit 20 prohibits the artificial intelligence from learning the learning data. For example, the time period during which the learning data is not disclosed to the public can be set as the prohibited time period, and the time after the learning data is disclosed to the public can be set as the permitted time period. According to this setting, the artificial intelligence is permitted to learn the learning data after the learning data is disclosed to the public, and the artificial intelligence is prohibited from learning the learning data while the learning data is not disclosed to the public. Needless to say, the permission period can be set for other purposes. The permission period can be set by a user with permission to set the permission period. For example, when creating learning data, attribute information including permission period information can be attached to the learning data. This processing can be performed by the attachment unit 26.

[0087] The changing unit 20 can be configured to change the learning process of the artificial intelligence according to attribute information attached to the learning data that indicates a limit on the number of times the learning data can be learned. The attribute information attached to the learning data may include number of times information indicating a limit on the number of times the learning data can be learned. The controller 28 counts the number of times the artificial intelligence has learned the learning data and attaches information indicating this number of times to the learning data. If the number of times the learning data has been learned exceeds the limit, the learning data is prohibited from being learned. That is, even when the user instructs the controller 28 to have the artificial intelligence learn the learning data, the controller 28 does not cause the artificial intelligence to learn the learning data. The controller 28 can count the number of times different artificial intelligences have learned, either cumulatively or individually. When the number of times different artificial intelligences have learned is counted cumulatively, learning data whose number of times of learning exceeds the limit is prohibited from being learned by any artificial intelligence. The limit can be set by a user with permission to set a limit on the number of times of learning. Note that a user with permission to set a limit on the number of times of learning can reset the counter for learning data whose number of times of learning exceeds the limit to 0. For example, when the learning data is created, attribute information including the number of times information is attached to the learning data. This processing can be performed by the attaching unit 26.

[0088] The changing unit 20 can be configured to change the AI ​​learning process according to attribute information indicating a specific AI attached to the learning data. The attribute information attached to the learning data can include information indicating AIs that are permitted to learn the learning data and / or AIs that are prohibited from learning the learning data. For example, the attribute information can indicate one or more predetermined AIs that are permitted to learn the learning data. As another example, the attribute information can indicate one or more predetermined devices that the AI ​​is permitted to learn the learning data. The one or more predetermined devices can be specific devices (e.g., information processing device 10, specific personal computers) or devices at specific facilities. When the learning data is related to the medical field, for example, devices in a hospital or medical school can be set as predetermined devices. AIs installed in devices other than those permitted are prohibited from learning the learning data. Since data related to the medical field often includes personal information, there may be situations where it is desirable to permit only limited AIs to learn the learning data and prohibit AIs other than those limited AIs from learning the learning data. This can prevent or suppress the leakage of personal information. This exemplary use in the medical field is merely an example, and the present disclosure can be applied to other fields. The AIs permitted to learn the learning data can be set by a user with permission to make this setting. When learning data is created, for example, attribute information including information indicating that artificial intelligence is permitted to learn the learning data is appended to the learning data. This processing can be performed by the appending unit 26.

[0089] The changing unit 20 can be configured to change the learning process of the artificial intelligence according to attribute information indicating a learning method attached to the learning data. The attribute information attached to the learning data may include information indicating learning methods permitted for learning the learning data and / or learning methods prohibited for learning the learning data. An example of an artificial intelligence learning method is the algorithm used in the artificial intelligence. Sometimes, suitable learning data for learning varies depending on the artificial intelligence algorithm. For example, when an artificial intelligence learns learning data, the performance of the artificial intelligence may improve in some cases and decrease in others depending on the artificial intelligence algorithm. In addition, the performance of the artificial intelligence may remain unchanged in some cases. Learning data that improves the performance of the artificial intelligence as a result of learning is learning data suitable for the artificial intelligence to learn. An artificial intelligence that improves its performance as a result of learning the learning data is an artificial intelligence that is permitted to learn the learning data. Furthermore, an artificial intelligence that does not improve its performance as a result of learning the learning data (i.e., an artificial intelligence whose performance decreases or whose performance remains unchanged) is an artificial intelligence that is prohibited from learning the learning data. For example, whether the performance of each artificial intelligence is improved as a result of learning varies depending on the learning history of the artificial intelligence learning the learning data. The attribute information attached to the learning data includes information indicating an algorithm of an artificial intelligence that is permitted to learn the learning data, and information indicating an algorithm of an artificial intelligence whose performance is improved as a result of learning the learning data. When, for example, learning data is created, attribute information including information indicating an algorithm of an artificial intelligence whose performance is improved as a result of learning the learning data is attached to the learning data. This processing can be performed by the attaching unit 26. In the case where the algorithm of the artificial intelligence attempting to learn the learning data is an algorithm of artificial intelligence whose performance is improved as a result of learning the learning data, the changing unit 20 permits the artificial intelligence to learn the learning data. In the case where the algorithm of the artificial intelligence attempting to learn the learning data is an algorithm of artificial intelligence whose performance is not improved as a result of learning the learning data, the changing unit 20 prohibits the artificial intelligence from learning the learning data. This enables the artificial intelligence to learn learning data that improves the performance of the artificial intelligence. Learning data that does not improve its performance as a result of learning can be set as learning data that the artificial intelligence is permitted to learn.

[0090] The change unit 20 can be configured to change the AI's learning process based on attribute information indicating a domain of use attached to the learning data. The attribute information attached to the learning data can include information indicating permitted domains for learning and / or prohibited domains for use of the learning data. Examples of domains of use include, but are not limited to, medicine, education, character recognition, translation, and business. For example, if information indicating the medical domain is included in the attribute information as information indicating permitted domains for learning, the learning data attached with the attribute information is permitted for learning by an AI used in the medical field. If the AI ​​attempting to learn the learning data is used in the medical field, the change unit 20 permits the AI ​​to learn the learning data. If the AI ​​attempting to learn the learning data is not used in the medical field, the change unit 20 prohibits the AI ​​from learning the learning data. This enables the AI ​​to learn learning data appropriate for the AI's domain of use. For example, when creating learning data, attribute information including information indicating permitted domains for learning the learning data is attached to the learning data. This process can be performed by the attaching unit 26. Information indicating multiple domains can be included in the attribute information. In this case, an AI used in at least one of the multiple domains can be permitted to learn the learning data. Alternatively, artificial intelligence used in all multiple fields may be permitted to learn learning data.

[0091] The changing unit 20 can allow or prohibit the artificial intelligence from learning the learning data according to the user who uses the learning data. For example, the attribute information attached to the learning data indicates the user identification information of the user who is permitted to make the artificial intelligence learn the learning data. In the case where the user identification information of the user who has been instructed to make the artificial intelligence learn the learning data is included in the attribute information attached to the learning data, the changing unit 20 allows the artificial intelligence to learn the learning data. In the case where the user identification information of the user who has been instructed to make the artificial intelligence learn the learning data is not included in the attribute information attached to the learning data, the changing unit 20 prohibits the artificial intelligence from learning the learning data. When creating learning data, for example, attribute information including the user identification information of the user who is permitted to make the artificial intelligence learn the learning data is attached to the learning data. This processing can be performed by the attaching unit 26.

[0092] In a first exemplary embodiment, in the case where artificial intelligence learning of learning data is prohibited, the controller 28 may send information to the creator of the learning data to inquire whether artificial intelligence learning of learning data is permitted. For example, the attribute information attached to the learning data may include information indicating the email address of the creator of the learning data or the address of the device used by the creator. The controller 28 sends the query information to the email address or the address of the device. In the case where the creator permits artificial intelligence learning of learning data, information indicating permission is sent to the information processing device 10 via email, or is sent from the device used by the creator to the information processing device 10. In the case where the creator permits, the change unit 20 permits artificial intelligence learning of learning data. In the case where the creator prohibits or does not permit artificial intelligence learning of learning data, information indicating prohibition is sent to the information processing device 10 via email, or is sent from the device used by the creator to the information processing device 10. In this case, the change unit 20 prohibits artificial intelligence learning of learning data.

[0093] In the first exemplary embodiment, receiving unit 18 may receive learning data to which attribute information indicating that artificial intelligence learning of the learning data is prohibited is attached. If the learning data received by receiving unit 18 does not have attribute information indicating that artificial intelligence learning of the learning data is prohibited, changing unit 20 permits artificial intelligence learning of the learning data. In this example, changing unit 20 is considered to be an example of a prohibiting unit.

[0094] In another example, receiving unit 18 may receive learning data with attribute information indicating that artificial intelligence is permitted to learn the learning data. If the learning data received by receiving unit 18 does not have attribute information indicating that artificial intelligence is permitted to learn the learning data, changing unit 20 prohibits artificial intelligence from learning the learning data. In this example, changing unit 20 is considered to be an example of a permitting unit.

[0095] The operations of the changing unit 20, the output unit 22, the deleting unit 24, and the adding unit 26 in the exemplary embodiment can be performed by artificial intelligence. For example, artificial intelligence instructed to learn learning data can perform the operations of these units in the exemplary embodiment.

[0096] Second Exemplary Embodiment

[0097] Will refer to Figure 4 An information processing apparatus according to a second exemplary embodiment of the present disclosure will be described. Figure 4 An example of the information processing apparatus according to the second exemplary embodiment is illustrated.

[0098] The information processing device 50 according to the second exemplary embodiment is configured to classify learning data of artificial intelligence learning as influential learning data or non-influence learning data. The information processing device 50 can classify learning data that has an impact on the performance of the artificial intelligence as influential learning data and classify learning data that has no impact on the performance of the artificial intelligence as non-influence learning data, and record the learning data in a recording unit in a manner that allows the influential learning data to be distinguished from the non-influence learning data.

[0099] The information processing device 50 may classify the influential learning data learned by the artificial intelligence as improving learning data or degrading learning data. The information processing device 50 may classify the influential learning data that improves the performance of the artificial intelligence as improving learning data and classify the influential learning data that degrades the performance of the artificial intelligence as degrading learning data, and record the influential learning data in the recording unit in a manner that allows the improving learning data to be distinguished from the degrading learning data.

[0100] The information processing device 50 may classify learning data learned by the artificial intelligence as improved learning data or non-improved learning data. The information processing device 50 may classify learning data that improves the performance of the artificial intelligence as improved learning data and classify learning data that degrades the performance of the artificial intelligence as non-improved learning data, and record the learning data in the recording unit in a manner that allows the improved learning data to be distinguished from the non-improved learning data.

[0101] The information processing device 50 is, for example, a PC, a tablet PC, a smartphone, a mobile phone, or any other type of device (eg, a multifunction printer having functions such as a print function). Needless to say, devices other than these may be the information processing device 50 .

[0102] As in the first exemplary embodiment, the algorithm used in artificial intelligence is not particularly limited, and the algorithm described in the first exemplary embodiment may be used.

[0103] Learning data is data for artificial intelligence to learn. As in the first exemplary embodiment, the learning data can be learning data for supervised learning or learning data for unsupervised learning. The type, data format, and content of the learning data are not particularly limited.

[0104] Artificial intelligence may be installed in the information processing device 50, or may be installed in a device other than the information processing device 50 (for example, a server or a PC). That is, a program for realizing artificial intelligence may be stored in the information processing device 50, or may be stored in a device other than the information processing device 50.

[0105] The configuration of the information processing apparatus 50 is described below in detail.

[0106] The communication unit 52 is a communication interface and has a function of sending information to another device and a function of receiving information from another device. The communication unit 52 may have a wireless communication function or a wired communication function. The communication unit 52 can communicate with another device via a communication path using wireless communication or wired communication. The communication path is, for example, a network such as a LAN or the Internet. The communication unit 52 can communicate with another device without a communication path by using, for example, short-range wireless communication. Short-range wireless communication is, for example, Bluetooth (registered trademark), RFID, or NFC.

[0107] For example, in the case where learning data is transmitted from a device other than the information processing device 50 to the information processing device 50, the communication unit 52 receives the learning data. In addition, the communication unit 52 may transmit the learning data to another device.

[0108] The UI unit 54 is a user interface. Examples include a display and an operation unit. The display is a display device such as a liquid crystal display. The operation unit is an input unit such as a keyboard, input keys, or an operation panel. A touch panel that functions as both a display and an operation unit is also an example of the UI unit 54.

[0109] The storage unit 56 is one or more storage areas storing various types of information. Each storage area is composed of, for example, one or more storage devices (e.g., physical drives such as hard disk drives or memories) provided in the information processing device 50. Learning data can be stored in the storage unit 56.

[0110] The receiving unit 58 is configured to receive learning data. For example, when the communication unit 52 receives learning data transmitted from a device other than the information processing device 50, the receiving unit 58 receives the learning data received by the communication unit 52. In addition, when the learning data is supplied to the information processing device 50 through the UI unit 54, or when the learning data is supplied to the information processing device 50 by using a storage device (e.g., a hard disk drive, a USB memory, a CD, a DVD, or any other portable storage medium), the receiving unit 58 receives the supplied learning data.

[0111] The determination unit 60 is configured to determine whether the learning data learned by the artificial intelligence has an impact on the performance of the artificial intelligence, and classify the learning data as influential learning data that has an impact on the performance of the artificial intelligence or non-influence learning data that has no impact on the performance of the artificial intelligence. This clearly indicates whether the learning data has an impact on the performance of the artificial intelligence. The learning data classified as influential learning data or non-influence learning data can be recorded in the recording unit.

[0112] The concept of influential learning data covers both improving learning data and degrading learning data. Improving learning data is learning data that improves the performance of the artificial intelligence that learned the learning data. Degrading learning data is learning data that degrades the performance of the artificial intelligence that learned the learning data. Non-influential learning data is learning data that does not change the performance of the artificial intelligence that learned the learning data. This clearly indicates whether the learning data improves or degrades the performance of the artificial intelligence. When the difference between the performance of the artificial intelligence before learning the learning data and the performance after learning the learning data is less than a threshold, the determination unit 60 can classify the learning data as non-influential learning data.

[0113] The determining unit 60 may classify the influential learning data learned by the artificial intelligence as improving learning data or degrading learning data. The learning data classified as improving learning data or degrading learning data may be recorded in the recording unit.

[0114] The performance of artificial intelligence is a performance related to the functions of artificial intelligence. Examples of the functions of artificial intelligence include, but are not limited to, character recognition, image recognition, speech recognition, object recognition, translation, creativity (e.g., creativity in fields such as business or art), and problem solving (the ability to solve problems in fields such as business).

[0115] The determining unit 60 may classify the learning data into influential learning data or non-influential learning data for each artificial intelligence. In addition, the determining unit 60 may classify the learning data into improving learning data or degrading learning data for each artificial intelligence.

[0116] The determining unit 60 may classify the learning data into impact learning data or non-impact learning data for each function of artificial intelligence. In addition, the determining unit 60 may classify the learning data into improvement learning data or reduction learning data for each function of artificial intelligence.

[0117] The determination unit 60 performs a test to determine whether the learning data has an impact on the performance of the artificial intelligence service. For example, the determination unit 60 causes the artificial intelligence to learn the learning data and determines whether the performance of the artificial intelligence has changed before and after the learning.

[0118] To determine whether the learning data has an impact on the performance of the character recognition function, the determination unit 60 may perform the following operations. The determination unit 60 first provides predetermined document data (test data) to an artificial intelligence that has not yet learned the learning data. The determination unit 60 then causes the artificial intelligence to recognize characters included in the test data and calculates the character recognition accuracy (e.g., character recognition rate).

[0119] Next, the determination unit 60 causes the artificial intelligence to learn the learning data. Then, the determination unit 60 provides the test data to the artificial intelligence that has learned the learning data, causes the artificial intelligence to recognize characters included in the document data, and calculates the character recognition accuracy.

[0120] The determination unit 60 determines whether the character recognition accuracy has changed before and after the learning. In the case where the character recognition accuracy has changed before and after the learning, the determination unit 60 determines that the learning data has an influence on the performance of the artificial intelligence regarding the character recognition function, and in the case where the character recognition accuracy has not changed before and after the learning, the determination unit 60 determines that the learning data has no influence on the performance of the artificial intelligence regarding the character recognition function. In the case where the character recognition accuracy has changed before and after the learning, the determination unit 60 classifies the learning data as influential learning data, and in the case where the character recognition accuracy has not changed before and after the learning, the determination unit 60 classifies the learning data as non-influence learning data.

[0121] Furthermore, in a case where the character recognition accuracy after learning is higher than the character recognition accuracy before learning, the determination unit 60 determines that the learning data is learning data that improves the performance of the artificial intelligence regarding the character recognition function, and in a case where the character recognition accuracy after learning is lower than the character recognition accuracy before learning, the determination unit 60 determines that the learning data is learning data that reduces the performance of the artificial intelligence regarding the character recognition function. In a case where the character recognition accuracy after learning is higher than the character recognition accuracy before learning, the determination unit 60 classifies the learning data as improved learning data, and in a case where the character recognition accuracy after learning is lower than the character recognition accuracy before learning, the determination unit 60 classifies the learning data as non-improved learning data.

[0122] Similarly, the determination unit 60 determines whether the learning data has an impact on the performance of functions other than character recognition of the artificial intelligence. For example, before and after the artificial intelligence learns the image learning data, the image recognition accuracy (e.g., accurate recognition rate) of the image test data is calculated, and the image learning data is classified based on the calculation result. For example, if the artificial intelligence that cannot recognize a cat in an image learns learning data including an image of a cat and becomes able to recognize a cat in the image, this means that the performance of the artificial intelligence with respect to the image recognition function is improved due to learning. On the other hand, if the artificial intelligence still cannot recognize the cat in the image even after learning, this means that the performance of the artificial intelligence with respect to the image recognition function is not improved due to learning.

[0123] The determination unit 60 can determine whether the learning data learned by the artificial intelligence improves the performance of the artificial intelligence, and classify the learning data as improved learning data that improves the performance of the artificial intelligence or non-improved learning data that does not improve the performance of the artificial intelligence. This clearly shows whether the learning data improves the performance of the artificial intelligence. The learning data classified as improved learning data or non-improved learning data can be recorded in the recording unit. When the difference between the performance of the artificial intelligence before learning the learning data and the performance of the artificial intelligence after learning the learning data is less than a threshold, the determination unit 60 can classify the learning data as non-improved learning data. The determination unit 60 can determine for each artificial intelligence whether the learning data learned by the artificial intelligence improves the performance of the artificial intelligence, and classify the learning data as improved learning data or non-improved learning data for each function of the artificial intelligence.

[0124] Note that the determination unit 60 may make the determination with respect to learning data specified by the user or specified in advance.

[0125] The appending unit 62 is configured to append determination result information to the learning data. The determination result information indicates the result of the determination made by the determination unit 60 with respect to the learning data. The appending unit 62 may append the determination result information to the learning data as accompanying information, or may embed the determination result information into the learning data itself. Note that the appending unit 62 may append attribute information according to the first exemplary embodiment to the learning data.

[0126] The controller 64 is configured to control the operation of each unit of the information processing device 50. The controller 64 includes a recording controller 66 and a learning controller 68.

[0127] The recording controller 66 is configured to record the learning data with the determination result information attached thereto in a recording unit. For example, the recording controller 66 classifies the learning data as influential learning data or non-influential learning data, and then records the learning data in the recording unit. The recording controller 66 may classify the influential learning data as improved learning data or reduced learning data, and then record the influential data in the recording unit. The recording controller 66 may classify the learning data as improved learning data or non-improved learning data, and then record the learning data in the recording unit. The recording unit in which the learning data is recorded may be provided in the information processing device 50, or may be provided in a device other than the information processing device 50 (e.g., a server or a PC). The recording unit in which the learning data is recorded may be specified by the user or may be predetermined.

[0128] The recording controller 66 may record learning history information indicating the learning history of the AI ​​in association with the learning data learned by the AI ​​in the recording unit. The learning history information may be attached to the learning data. The learning history information associated with the learning data indicates the previous learning data learned by the AI ​​that has learned the learning data. The learning history information may indicate the correspondence between the AI ​​and the learning data learned by the AI ​​in the past. Furthermore, the learning history information may indicate the order in which the AI ​​learned the learning data sets. For example, learning data identification information used to identify the learning data learned by the AI ​​in the past may be associated with the AI, and the recording controller 66 may associate the learning data identification information associated with the AI ​​that learned the learning data with the learning data. For example, if AI α sequentially learned learning data sets A, B, and C, information indicating that AI α sequentially learned learning data sets A, B, and C is associated with learning data set C as learning history information. The term "learning data set" is used here for convenience and is not intended to be limiting. A learning data set may include multiple learning data blocks, or may be just one learning data block. Examples of learning data blocks include images, documents, videos, and the like.

[0129] The learning controller 68 is configured to cause the artificial intelligence to learn learning data. For example, the learning controller 68 causes the artificial intelligence to learn influential learning data or non-influential learning data that the artificial intelligence has not yet learned, based on the learning purpose of the artificial intelligence. More specifically, the learning controller 68 causes the second artificial intelligence to learn influential learning data or non-influential learning data classified according to the learning results of the first artificial intelligence, based on the purpose of the second artificial intelligence. The first artificial intelligence and the second artificial intelligence are different artificial intelligences. Specifically, the learning controller 68 causes the second artificial intelligence to learn improving learning data or degrading learning data classified according to the learning results of the first artificial intelligence, based on the learning purpose of the second artificial intelligence. The learning controller 68 can cause the second artificial intelligence to learn improving learning data or non-improving learning data classified according to the learning results of the first artificial intelligence, based on the learning purpose of the second artificial intelligence. The second artificial intelligence can be specified, for example, by the user or pre-specified.

[0130] The processing of learning controller 68 will be described using a specific example. For example, if the performance of function a of artificial intelligence α improves as a result of learning learning dataset A, and another artificial intelligence β is learning to improve function a of artificial intelligence β, learning controller 68 causes artificial intelligence β to learn learning dataset A. Specifically, since learning dataset A is classified as improved learning data capable of improving the performance of function a, learning controller 68 causes artificial intelligence β to learn learning dataset A, which is improved learning data that improves the performance of function a. In other words, if the learning goal of artificial intelligence β is to bring the performance of artificial intelligence β closer to that of artificial intelligence α, artificial intelligence β is caused to learn learning dataset A, which improves the performance of artificial intelligence α.

[0131] Furthermore, if the purpose of learning artificial intelligence β is to prevent the performance of artificial intelligence β from approaching that of artificial intelligence α (i.e., to make the performance of artificial intelligence β different from that of artificial intelligence α), learning controller 68 causes artificial intelligence β to learn degrading learning data that degrades the performance of artificial intelligence α, learning data that has no effect learned by artificial intelligence α, or learning data that has no improvement learned by artificial intelligence α. In this way, artificial intelligence β becomes an artificial intelligence with performance different from that of artificial intelligence α.

[0132] For example, the user specifies the learning purpose of the artificial intelligence. For example, the user specifies the learning data and the learning purpose of the artificial intelligence (e.g., artificial intelligence β) by operating the UI unit 54. The learning controller 68 causes the artificial intelligence specified by the user to learn the learning data that matches the learning purpose specified by the user.

[0133] The learning controller 68 can cause the second artificial intelligence to learn the influential learning data that was classified as a result of the learning of the first artificial intelligence whose function corresponds to the function of the second artificial intelligence. The function corresponding to the function of the second artificial intelligence can be the same function as the function of the second artificial intelligence, or can be a function similar to the function of the second artificial intelligence (for example, a function whose difference from the function of the second artificial intelligence is equal to or less than a threshold value).

[0134] For example, if the performance of function a of artificial intelligence α is improved as a result of artificial intelligence α learning learning data set A, learning controller 68 causes artificial intelligence β, which also has function a, to learn learning data set A. In other words, learning controller 68 causes artificial intelligence β to learn the improved learning data that improves the performance of artificial intelligence α. Artificial intelligence α and artificial intelligence β can be specified by the user or pre-specified, for example.

[0135] In addition, the learning controller 68 can cause the second artificial intelligence to learn influential learning data that has been classified as a result of the learning history of the first artificial intelligence corresponding to the learning history of the second artificial intelligence. The learning history corresponding to the learning history of the second artificial intelligence can be the same learning history as the learning history of the second artificial intelligence, or can be a learning history similar to the learning history of the second artificial intelligence (for example, a learning history that differs from the learning history of the second artificial intelligence by a threshold or less). The learning history is the history of the artificial intelligence learning learning data in the past.

[0136] Below, refer to Figures 5 to 7 The processing of the information processing device 50 will be described. Figures 5 to 7 An example of learning data according to the second exemplary embodiment is illustrated.

[0137] like Figure 5 As illustrated in FIG, a learning data set A is designated as the learning data to be determined, and the artificial intelligence α( Figure 5 The AI ​​(α) in the figure is designated as the artificial intelligence to learn the learning dataset A. For example, when the user gives an instruction to determine whether the learning dataset A is influential learning data or non-influence learning data by operating the UI unit 54, the determination unit 60 causes the artificial intelligence α to learn the learning dataset A and performs a test to determine whether the learning dataset A has an impact on the performance of the artificial intelligence α. That is, the determination unit 60 determines whether the performance of the artificial intelligence α has changed before and after learning the learning dataset A. If the performance of the artificial intelligence α has changed, the determination unit 60 determines that the learning dataset A is influential learning data that has an impact on the performance of the artificial intelligence α. If the performance of the artificial intelligence α has not changed, the determination unit 60 determines that the learning dataset A is non-influence learning data that has no impact on the performance of the artificial intelligence α.

[0138] In the case where the learning dataset A is classified as influential learning data, the appending unit 62 appends determination result information 70 indicating that the learning dataset A is influential learning data to the learning dataset A, such as Figure 5 In the case where the learning dataset A is classified as non-influence learning data, the appending unit 62 appends determination result information 72 indicating that the learning dataset A is non-influence learning data to the learning dataset A, such as Figure 5 exemplified in .

[0139] The recording controller 66 may cause the learning data set A classified as influential learning data or non-influential learning data to be recorded in the recording unit.

[0140] The determination unit 60 can determine whether the performance of the artificial intelligence α is improved before and after learning the learning dataset A. In the case where the performance of the artificial intelligence α after learning the learning dataset A is improved relative to the performance of the artificial intelligence α before learning the learning dataset A, the determination unit 60 determines that the learning dataset A is improved learning data that improves the performance of the artificial intelligence α. In this case, Figure 6 As illustrated in FIG, the appending unit 62 appends determination result information 74 indicating that the learning dataset A is improved learning data to the learning dataset A. In a case where the performance of the artificial intelligence α after learning the learning dataset A is reduced relative to the performance of the artificial intelligence α before learning the learning dataset A, the determining unit 60 determines that the learning dataset A is reduced learning data that reduces the performance of the artificial intelligence α. In this case, as Figure 6 As illustrated in , the appending unit 62 appends, to the learning dataset A, determination result information 76 indicating that the learning dataset A is reduced learning data.

[0141] The recording controller 66 may cause the learning data set A classified as improvement learning data or reduction learning data to be recorded in the recording unit.

[0142] In addition, in the case where the learning data set A is classified as improved learning data, the learning controller 68 may enable another designated artificial intelligence ( Figure 6 The AI ​​(β) in the learning controller 68 learns the learning data set A. This can improve the performance of the artificial intelligence β. In addition, if the learning data set A is classified as degraded learning data, the learning controller 68 can prohibit the artificial intelligence β from learning the learning data set A. This prevents the performance of the artificial intelligence β from being degraded.

[0143] In the case where the performance of artificial intelligence α after learning the learning dataset A does not improve relative to the performance of artificial intelligence α before learning the learning dataset A, the determination unit 60 may determine that the learning dataset A is non-improvement learning data that does not improve the performance of artificial intelligence α. In the case where the performance of artificial intelligence α after learning the learning dataset A improves relative to the performance of artificial intelligence α before learning the learning dataset A, the determination unit 60 determines that the learning dataset A is improvement learning data that improves the performance of artificial intelligence α. In the case where the learning dataset A is classified as improvement learning data, the appending unit 62 appends determination result information 78 indicating that the learning dataset A is improvement learning data to the learning dataset A, as shown in FIG. Figure 7 In the case where the learning dataset A is classified as no-improvement learning data, the appending unit 62 appends the determination result information 80 indicating that the learning dataset A is no-improvement learning data to the learning dataset A, such as Figure 7 exemplified in .

[0144] The recording controller 66 may cause the learning data set A classified as the improvement learning data or the non-improvement learning data to be recorded in the recording unit.

[0145] Furthermore, in the case where the learning data set A is classified as improved learning data, the learning controller 68 may enable another artificial intelligence ( Figure 7 The AI ​​(β) in the training data set A is used to learn the training data set A. This improves the performance of the artificial intelligence β. In addition, if the training data set A is classified as non-improving learning data, the learning controller 68 prohibits the artificial intelligence β from learning the training data set A. This prevents the performance of the artificial intelligence β from being degraded.

[0146] The determination unit 60 can determine the influence of the learning data on the artificial intelligence for each function of each artificial intelligence and create management information (e.g., a database) for managing the determination results. The management information can be stored in the storage unit 56 or in a device other than the information processing device 50.

[0147] Figure 8 An example of a database is illustrated as an example of management information. Figure 8 The database illustrated in is a database that stores determination results indicating the influence of the learning dataset A on artificial intelligence that has learned the learning dataset A. In this database, the determination results indicate the influence of the learning dataset A on each function of the artificial intelligence.

[0148] For example, AI α ( Figure 8 AI (α)) and artificial intelligence β ( Figure 8The AI ​​(β) in each has more functions, including character recognition, translation, creativity, and problem solving. Determination result A indicates that the performance has been significantly improved. Determination result B indicates that the performance has been slightly improved. Determination result C indicates that the performance has not changed. Determination result D indicates that the performance has been reduced.

[0149] As a result of learning data set A, AI α's character recognition rate significantly improved, AI α's translation accuracy slightly improved, AI α's creativity remained unchanged, and AI α's problem-solving ability decreased. Specifically, AI α's character recognition performance significantly improved, AI α's translation performance slightly improved, AI α's creativity remained unchanged, and AI α's problem-solving performance decreased.

[0150] As a result of learning dataset A, AI β's character recognition rate, translation accuracy, and creativity did not change, but AI β's problem-solving ability decreased. That is, AI β's character recognition function, translation function, and creativity performance did not change, but AI β's problem-solving performance decreased.

[0151] Since the determination results are managed as described above, it is possible to evaluate the influence of the learning data on the performance of each artificial intelligence. Figure 8 In the example illustrated in , the comparison between artificial intelligence α and artificial intelligence β shows that, despite the fact that both artificial intelligence α and artificial intelligence β learned the same learning data set A, the improvement in the performance of artificial intelligence β was not as great as the improvement in the performance of artificial intelligence α. In other words, the comparison shows that the improvement in the performance of artificial intelligence α was greater than the improvement in the performance of artificial intelligence β.

[0152] Because AI α and AI β often have different learning histories, it may still be difficult to generalize the impact of learning dataset A on the performance of AI based solely on these determination results. However, these determination results can provide a way to evaluate the impact of learning dataset A on different AIs.

[0153] Furthermore, when managing the learning history of an AI, it is possible to estimate the learning history that is likely to improve the performance of the AI ​​when learning a specific dataset. For example, if AIs α and β, each with a different learning history, use the same algorithm and learn the same learning dataset A, if the effect of learning dataset A on AI α differs from that on AI β, this difference may result from their different learning histories. This makes it possible to estimate the learning history that is likely to improve the performance of the AI ​​when learning dataset A.

[0154] Furthermore, in the case where artificial intelligence α and artificial intelligence β use different algorithms, it is possible to estimate an algorithm that is likely to improve the performance of artificial intelligence when artificial intelligence learns a specific data set. Figure 8 In the example illustrated in , the performance of artificial intelligence α is improved more by learning learning dataset A than the performance of artificial intelligence β is improved by learning learning dataset A. Therefore, it can be estimated that the algorithm used by artificial intelligence α can explain the improvement in the performance of artificial intelligence α when artificial intelligence α learns learning dataset A.

[0155] In addition, the database can manage the algorithm used by each AI, the time when the AI ​​started using the AI, the time when the AI ​​started learning, and the learning period of the AI, etc. Sometimes, one or more factors may affect the performance of the AI ​​learning a specific learning data set. When managing these details, it is possible to estimate whether the impact of learning a specific data set on the performance of the AI ​​(for example, improving or reducing performance) is due to the learning data itself or other factors (for example, the algorithm or learning history of the AI).

[0156] Hereinafter, modifications of the second exemplary embodiment are described.

[0157] First Modification of the Second Exemplary Embodiment

[0158] Below, a first modification of the second exemplary embodiment is described. In the first modification of the second exemplary embodiment, the determination unit 60 causes the artificial intelligence to learn a combination of learning data sets and determines whether the combination has an impact on the performance of the artificial intelligence. The determination unit 60 then classifies the combination as a combination that has an impact on the performance of the artificial intelligence or a combination that has no impact on the performance of the artificial intelligence. The term "learning data set" is used here for convenience and is not intended to be limiting. The learning data set may include multiple learning data blocks, or may be just one learning data block. Examples of learning data blocks include images, documents, videos, and the like.

[0159] The determination unit 60 may determine whether the combination of the learning data sets improves the performance of artificial intelligence, and classify the combination as a combination that improves the performance of artificial intelligence or a combination that degrades the performance of artificial intelligence.

[0160] The determination unit 60 may determine whether the combination of the learning data sets improves the performance of artificial intelligence, and classify the combination as a combination that improves the performance of artificial intelligence or a combination that does not improve the performance of artificial intelligence.

[0161] The adding unit 62 adds determination result information indicating the determination result of the determining unit 60 to the combination of learning data sets. The recording controller 66 may cause the combination to which the determination result information is added to be recorded in the recording unit.

[0162] Allowing artificial intelligence to learn the combination of learning data sets means providing the learning data sets to artificial intelligence at the same time and allowing the artificial intelligence to learn the combination of the learning data sets, or providing the combined learning data sets to artificial intelligence separately and allowing the artificial intelligence to learn the learning data sets separately.

[0163] Below, refer to Figures 9 to 11 The process according to the first modification of the second exemplary embodiment is described. Figures 9 to 11 An example of learning data according to the first modification of the second exemplary embodiment is illustrated.

[0164] like Figure 9 As illustrated in , for example, a combination of a learning data set A and a learning data set B is designated as learning data, and artificial intelligence α( Figure 9 The AI ​​(α) in the figure is designated as the artificial intelligence that learns the combination of learning dataset A and learning dataset B. For example, when the user gives an instruction to determine the combination of learning dataset A and learning dataset B by operating the UI unit 54, the determination unit 60 causes the artificial intelligence α to learn the combination of learning dataset A and learning dataset B, and performs a test to determine whether the combination of learning dataset A and learning dataset B has an impact on the performance of the artificial intelligence α. That is, the determination unit 60 determines whether the performance of the artificial intelligence α changes before and after learning the combination of learning dataset A and learning dataset B. If the performance of the artificial intelligence α changes, the determination unit 60 determines that the combination of learning dataset A and learning dataset B is a combination that has an impact on the performance of the artificial intelligence α. If the performance of the artificial intelligence α does not change, the determination unit 60 determines that the combination of learning dataset A and learning dataset B is a combination that has no impact on the performance of the artificial intelligence α.

[0165] In the case where the combination of the learning dataset A and the learning dataset B is classified as a combination that affects the performance of the artificial intelligence α, the appending unit 62 appends determination result information 82 indicating that the combination of the learning dataset A and the learning dataset B is a combination that has an influence on the performance of the artificial intelligence α to the combination of the learning dataset A and the learning dataset B, as shown in FIG. Figure 9 In the case where the combination of the learning dataset A and the learning dataset B is classified as a combination that has no influence on the performance of the artificial intelligence α, the appending unit 62 appends determination result information 84 indicating that the combination of the learning dataset A and the learning dataset B is a combination that has no influence on the performance of the artificial intelligence α to the combination of the learning dataset A and the learning dataset B, as shown in FIG. Figure 9 exemplified in .

[0166] The recording controller 66 may cause the combination of the learning data set A and the learning data set B classified as a combination having an influence on the performance of the artificial intelligence α or a combination having no influence on the performance of the artificial intelligence α to be recorded in the recording unit.

[0167] The determination unit 60 may determine whether the performance of the artificial intelligence α is improved before and after learning the combination of the learning dataset A and the learning dataset B. In a case where the performance of the artificial intelligence α after learning the combination of the learning dataset A and the learning dataset B is improved relative to the performance of the artificial intelligence α before learning the combination of the learning dataset A and the learning dataset B, the determination unit 60 determines that the combination of the learning dataset A and the learning dataset B is a combination that improves the performance of the artificial intelligence α. In this case, as Figure 10 As illustrated in FIG, the appending unit 62 appends determination result information indicating that the combination of the learning dataset A and the learning dataset B is a combination that improves the performance of the artificial intelligence α to the combination of the learning dataset A and the learning dataset B. In a case where the performance of the artificial intelligence α after learning the combination of the learning dataset A and the learning dataset B is reduced relative to the performance of the artificial intelligence α before learning the combination of the learning dataset A and the learning dataset B, the determining unit 60 determines that the combination of the learning dataset A and the learning dataset B is a combination that reduces the performance of the artificial intelligence α. In this case, as Figure 10 As illustrated in , the adding unit 62 adds, to the combination of the learning dataset A and the learning dataset B, determination result information 88 indicating that the combination of the learning dataset A and the learning dataset B is a combination that degrades the performance of the artificial intelligence α.

[0168] The recording controller 66 may cause the combination of the learning data set A and the learning data set B classified as a combination that improves the performance of the artificial intelligence α or a combination that degrades the performance of the artificial intelligence α to be recorded in the recording unit.

[0169] If the combination of learning dataset A and learning dataset B improves the performance of artificial intelligence α, the learning controller 68 may cause another designated artificial intelligence (e.g., artificial intelligence β) to learn the combination of learning dataset A and learning dataset B. Furthermore, if the combination of learning dataset A and learning dataset B degrades the performance of artificial intelligence α, the learning controller 68 may prohibit another artificial intelligence from learning the combination of learning dataset A and learning dataset B.

[0170] In a case where the performance of artificial intelligence α after learning the combination of learning dataset A and learning dataset B does not improve relative to the performance of artificial intelligence α before learning the combination of learning dataset A and learning dataset B, the determination unit 60 may determine that the combination of learning dataset A and learning dataset B is a combination that does not improve the performance of artificial intelligence α. In a case where the performance of artificial intelligence α after learning the combination of learning dataset A and learning dataset B improves relative to the performance of artificial intelligence α before learning the combination of learning dataset A and learning dataset B, the determination unit 60 determines that the combination of learning dataset A and learning dataset B is a combination that improves the performance of artificial intelligence α. In a case where the combination of learning dataset A and learning dataset B is classified as a combination that improves the performance of artificial intelligence α, the appending unit 62 appends determination result information 90 indicating that the combination of learning dataset A and learning dataset B is a combination that improves the performance of artificial intelligence α to the combination of learning dataset A and learning dataset B, as shown in FIG. Figure 11 In the case where the combination of the learning dataset A and the learning dataset B is classified as a combination that does not improve the performance of the artificial intelligence α, the appending unit 62 appends determination result information 92 indicating that the combination of the learning dataset A and the learning dataset B is a combination that does not improve the performance of the artificial intelligence α to the combination of the learning dataset A and the learning dataset B, as shown in FIG. Figure 11 exemplified in .

[0171] The recording controller 66 may cause the combination of the learning data set A and the learning data set B classified as a combination that improves the performance of the artificial intelligence α or a combination that does not improve the performance of the artificial intelligence α to be recorded in the recording unit.

[0172] In a case where the combination of the learning dataset A and the learning dataset B is classified as a combination that does not improve the performance of the artificial intelligence α, the learning controller 68 may prohibit another artificial intelligence from learning the combination of the learning dataset A and the learning dataset B.

[0173] The determination unit 60 may determine the influence of the combination of learning data sets on the AI ​​for each AI and create management information (e.g., a database) for managing the determination results. The management information may be recorded in the storage unit 56 or in a device other than the information processing device 10.

[0174] Figure 12 An example of a database is illustrated as an example of management information. Figure 12The database illustrated in the example is a database that stores determination results indicating the influence of the combination of learning dataset A and learning dataset B on the artificial intelligence that has learned the combination of learning dataset A and learning dataset B. In this database, the determination results indicate the influence of the combination of learning dataset A and learning dataset B on each function of the artificial intelligence. The meanings of the determination results A, B, C, and D are the same as those in Figure 8 The determination results exemplified in have the same meaning.

[0175] As a result of learning the combination of learning data sets A and B, the character recognition rate of artificial intelligence α significantly improved, the translation accuracy of artificial intelligence α slightly improved, the creativity of artificial intelligence α remained unchanged, and the problem-solving ability of artificial intelligence α decreased. In other words, the performance of the character recognition function of artificial intelligence α significantly improved, the performance of the translation function of artificial intelligence α slightly improved, the creativity of artificial intelligence α remained unchanged, and the problem-solving ability of artificial intelligence α decreased.

[0176] As a result of learning the combination of learning dataset A and learning dataset B, AI β's character recognition rate, translation accuracy, and creativity did not change, while AI β's problem-solving ability decreased. In other words, AI β's character recognition function, translation function, and creativity performance did not change, while AI β's problem-solving performance decreased.

[0177] Since the determination results are managed as described above, it is possible to evaluate the influence of the combination of learning data sets on each artificial intelligence. Figure 12 In the example illustrated in , the comparison between artificial intelligence α and artificial intelligence β shows that, despite the fact that artificial intelligence α and artificial intelligence β learned the same combination of learning data set A and learning data set B, the improvement in the performance of artificial intelligence β was not as great as the improvement in the performance of artificial intelligence α. In other words, this comparison shows that the improvement in the performance of artificial intelligence α was greater than the improvement in the performance of artificial intelligence β.

[0178] When AI learns learning datasets A and B separately in combination, the database can store information indicating the order in which learning datasets A and B were learned. This is because the order of learning can affect the performance of the AI. That is, the determination results obtained when the AI ​​learns learning dataset A and then learns dataset B, as well as the determination results obtained when the AI ​​learns learning dataset A and then learns dataset B, can both be managed in the database.

[0179] Although the determination unit 60 causes the artificial intelligence to learn a combination of two learning data sets in the above example, the determination unit 60 may cause the artificial intelligence to learn a combination of three or more learning data sets and determine the influence of the learning.

[0180] A combination of learning datasets may include only learning datasets of the same type or format, or may include learning datasets of different types or formats. For example, a combination of learning datasets may include a combination of multiple document datasets. Alternatively, a combination of learning datasets may include a combination of a document dataset and an image dataset. These combinations are merely examples, and other combinations are also possible. The learning datasets included in the combination may be specified by the user.

[0181] Second Modification of the Second Exemplary Embodiment

[0182] The following describes a second modification of the second exemplary embodiment. In this second modification, determination unit 60 determines the lifespan of the AI. As used herein, "lifespan" refers to a state in which the performance of the AI ​​has not improved by learning new learning data, or the improvement in AI performance achieved by learning new learning data falls below a threshold. Determining that the AI ​​has reached its lifespan is used as a criterion for prompting the user to replace the AI ​​or modify its algorithm.

[0183] For example, Figure 13 As illustrated in FIG, determining unit 60 causes artificial intelligence α to learn a learning dataset C that has not yet been learned by artificial intelligence α, and determines whether the performance of artificial intelligence α has improved as a result of the learning. If the performance of artificial intelligence α has improved, determining unit 60 determines that artificial intelligence α has not reached the end of its lifespan. If the performance of artificial intelligence α has not improved, determining unit 60 determines that artificial intelligence α has reached the end of its lifespan. Even if the performance of artificial intelligence α has improved, determining unit 60 may determine that artificial intelligence α has reached the end of its lifespan if the improvement is less than a threshold.

[0184] If it is determined that the artificial intelligence α has reached the end of its lifespan, the controller 64 may output information such as information indicating a recommendation to replace the artificial intelligence with another artificial intelligence or information indicating a recommendation to change the algorithm of the artificial intelligence. For example, the controller 64 may cause information indicating this recommendation to be displayed on the display of the UI unit 54, or may output a sound (e.g., voice, tone) indicating this recommendation from the speaker.

[0185] Note that the determination unit 60 may cause the artificial intelligence α to learn the same learning data a plurality of times continuously or at predetermined time intervals, and determine the life end of the artificial intelligence α based on the results of the plurality of learning rounds.

[0186] The determination unit 60 may cause the artificial intelligence α to learn multiple different learning data sets and determine the lifespan of the artificial intelligence α based on the learning results of the multiple different learning data sets. For example, if the number of learning data sets that improve the performance of the artificial intelligence α is less than a threshold, the determination unit 60 determines that the artificial intelligence α has reached the lifespan end. If the number of learning data sets that improve the performance of the artificial intelligence α is equal to or greater than the threshold, the determination unit 60 determines that the artificial intelligence α has not reached the lifespan end.

[0187] Figure 14 Another example is illustrated. In this example, determination unit 60 causes artificial intelligence α and artificial intelligence β, which have not yet learned learning dataset C, to learn learning dataset C. Determination unit 60 then compares the result of artificial intelligence α learning learning dataset C with the result of artificial intelligence β learning dataset C. If the difference between the learning result of artificial intelligence α and the learning result of artificial intelligence β is less than a threshold value, determination unit 60 determines that neither artificial intelligence α nor artificial intelligence β has reached the end of its lifespan. If the difference between the learning result of artificial intelligence α and the learning result of artificial intelligence β is equal to or greater than a threshold value, determination unit 60 determines that the artificial intelligence with the lower learning effect between artificial intelligence α and artificial intelligence β has reached the end of its lifespan. An example of an artificial intelligence with a lower learning effect is an artificial intelligence whose performance does not improve as much as that of other artificial intelligences that have learned the learning dataset. Another example is an artificial intelligence whose performance deteriorates more than that of other artificial intelligences that have learned the learning dataset.

[0188] Artificial intelligence α and artificial intelligence β may be artificial intelligences with the same or similar learning histories, or may be artificial intelligences with different or dissimilar learning histories. A case where the learning histories are similar is a case where the difference between the learning histories of artificial intelligence α and artificial intelligence β is less than a threshold.

[0189] The processing of the determination unit 60, the addition unit 62, the recording controller 66, and the learning controller 68 can be performed by artificial intelligence. For example, artificial intelligence instructed to learn learning data can perform these kinds of processing.

[0190] The functions of the units of the information processing device 10 and the information processing device 50 are realized, for example, by the collaboration of hardware and software. Specifically, the information processing device 10 and the information processing device 50 have one or more processors such as a CPU (not shown). The one or more processors read out and execute a program stored in a storage device (not shown), thereby realizing the functions of the units of the information processing device 10 and the information processing device 50. The program is stored in the storage device via a recording medium such as a CD or DVD or a communication path such as a network. In another example, the functions of the units of the information processing device 10 and the information processing device 50 can be realized by hardware resources such as a processor, an electronic circuit, and an application-specific integrated circuit (ASIC). A device such as a memory can be used to realize the functions of the units of the information processing device 10 and the information processing device 50. In yet another example, the functions of the units of the information processing device 10 and the information processing device 50 can be realized, for example, by a digital signal processor (DSP) or a field programmable gate array (FPGA).

[0191] The above description of exemplary embodiments of the present disclosure has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Obviously, many modifications and variations will be apparent to those skilled in the art. The embodiments have been selected and described to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to understand the various embodiments and modifications of the present disclosure as suitable for the specific use contemplated. The scope of the present disclosure is intended to be defined by the appended claims and their equivalents.

Claims

1. An information processing device comprising: a controller configured to cause a second artificial intelligence to learn, according to a learning objective of the second artificial intelligence, improvement learning data that improves the performance of the first artificial intelligence or reduction learning data that reduces the performance of the first artificial intelligence, which are classified according to a learning result of the first artificial intelligence; and to record the improvement learning data and the reduction learning data in a recording unit; in, In the case where the learning purpose of the second artificial intelligence is to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the controller causes the second artificial intelligence to learn the improved learning data, and Wherein, when the learning purpose of the second artificial intelligence is not to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the controller enables the second artificial intelligence to learn the reduced learning data.

2. An information processing device comprising a controller that causes the second artificial intelligence to learn, according to a learning objective of the second artificial intelligence, improved learning data that improves the performance of the first artificial intelligence or non-improvement learning data that does not improve the performance of the first artificial intelligence, as classified according to a learning result of the first artificial intelligence, and causes the improved learning data and the non-improvement learning data to be recorded in a recording unit. in, In the case where the learning purpose of the second artificial intelligence is to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the controller causes the second artificial intelligence to learn the improved learning data, and In which, when the learning purpose of the second artificial intelligence is not to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the controller enables the second artificial intelligence to learn the non-improvement learning data.

3. A non-transitory computer-readable medium storing a program for causing a computer to execute an image processing process, the process comprising causing a second artificial intelligence to learn, according to a learning objective of the second artificial intelligence, improvement learning data that improves the performance of the first artificial intelligence or reduction learning data that reduces the performance of the first artificial intelligence, which are classified according to the learning results of the first artificial intelligence, and causing the improvement learning data and the reduction learning data to be recorded in a recording unit. in, In the case where the learning purpose of the second artificial intelligence is to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the second artificial intelligence is made to learn the improved learning data, and Wherein, when the learning purpose of the second artificial intelligence is not to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the second artificial intelligence is made to learn the reduced learning data.

4. A non-transitory computer-readable medium storing a program for causing a computer to execute an image processing process, the process comprising causing a second artificial intelligence to learn, according to a learning objective of the second artificial intelligence, either improved learning data that improves the performance of the first artificial intelligence or non-improvement learning data that does not improve the performance of the first artificial intelligence, as categorized by a learning result of the first artificial intelligence, and causing the improved learning data and the non-improvement learning data to be recorded in a recording unit. in, In the case where the learning purpose of the second artificial intelligence is to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the second artificial intelligence is made to learn the improved learning data, and In which, when the learning purpose of the second artificial intelligence is not to make the performance of the second artificial intelligence close to the performance of the first artificial intelligence, the second artificial intelligence is made to learn the non-improvement learning data.

Images