Learning device, learning method, and learning program

Abstract

[Problem] To provide a learning device capable of preventing leakage of processing model information and easily starting up an inference environment by a processing model. [Solution] A learning device including: a learning data reception unit for receiving an input of learning data; a learning unit for generating a processing model through learning using the learning data; a configuration information acquisition unit for acquiring configuration information necessary for generating an inference system that executes inference processing on data; and an inference system generation unit for generating an inference system including the processing model configured from binary data, on the basis of the processing model generated by the learning unit and the configuration information acquired by the configuration information acquisition unit.

Description

Learning device, learning method, and learning program

[0001] The present invention relates to a learning device, a learning method, and a learning program.

[0002] In recent years, various technologies using machine learning have been developed. For example, a processing model obtained by machine learning can classify various types of data into multiple classes. The processing model extracts features from the data and classifies the data into one of multiple classes based on the extracted features. For example, at a product manufacturing site, a processing model that classifies input image data into good or bad products can be used for product inspection.

[0003] However, when the processing model is provided to a user who is a customer, there is a possibility that information about the processing model may be leaked to the user.

[0004] The following prior art is disclosed in the following patent document: A learning program that is obfuscated and includes an encryption key is executed. As a result, training data is applied to generate a processing model through machine learning, and model information included in the processing model is encrypted with the encryption key to generate encrypted model information. As a result, the learning program can be executed on-premises and the model information of the generated processing model can be output without the learning program and model information of the processing model being read by a user.

[0005] Japanese Patent Application Laid-Open No. 2021-174327

[0006] However, in the above prior art, there is a possibility that model information of the processing model may be leaked if the key is leaked from the learning program.

[0007] The present invention has been made to solve the above-mentioned problems. That is, an object of the present invention is to provide a learning device, a learning method, and a learning program that can prevent the leakage of processing model information and easily set up an inference environment using a processing model.

[0008] The above-mentioned problems of the present invention are solved by the following means.

[0009] (1) A learning device having a learning data accepting unit that accepts input of learning data, a learning unit that generates a processing model by learning using the learning data, a configuration information acquisition unit that acquires configuration information necessary for generating an inference system that performs inference processing on data, and an inference system generation unit that generates the inference system including the processing model consisting of binary data based on the processing model generated by the learning unit and the configuration information acquired by the configuration information acquisition unit.

[0010] (2) The learning device described in (1) above further comprises a key generation unit that generates an encryption key for encrypting the processing model generated by the learning unit and a decryption key for decrypting the encrypted processing model encrypted with the encryption key, and an encryption unit that generates the encrypted processing model by encrypting the processing model using the encryption key, wherein the inference system generation unit generates the inference system including the encrypted processing model, the decryption key, and the configuration information.

[0011] (3) The learning device described in (1) above further comprises a storage control unit that stores, in a storage area that is not accessible to the user, an encryption key for encrypting the processing model generated by the learning unit and a decryption key for decrypting the encrypted processing model encrypted with the encryption key, a key acquisition unit that acquires the encryption key and the decryption key from the storage area, and an encryption unit that generates the encrypted processing model by encrypting the processing model using the acquired encryption key, wherein the inference system generation unit generates the inference system including the encrypted processing model, the decryption key, and the configuration information.

[0012] (4) A learning device described in (1) or (3) above, further comprising a recommended pre-learning data acquisition unit that acquires recommended pre-learning data for pre-learning of the learning unit, and a learning control unit that performs pre-learning on the learning unit using the recommended pre-learning data.

[0013] (5) The learning device described in (4) above, wherein the recommended pre-training data includes at least one of pre-training data, pre-training setting parameters, and a pre-training model.

[0014] (6) The learning control unit selects and sets data to be pre-learned by the learning unit from at least one of pre-learning data, pre-learning setting parameters, and pre-learning models acquired by the recommended pre-learning data acquisition unit, in the learning device described in (5) above.

[0015] (7) A method for causing a learning device to execute the learning method, comprising: a step (a) of accepting input of learning data; a step (b) of generating a processing model by learning using the learning data; a step (c) of acquiring configuration information necessary for generating an inference system that performs inference processing on data; and a step (d) of generating the inference system including the processing model consisting of binary data based on the processing model generated in step (b) and the configuration information acquired in step (c).

[0016] (8) The learning method described in (7) above further comprises a step (e) of generating an encryption key for encrypting the processing model generated in step (b) and a decryption key for decrypting the encrypted processing model encrypted with the encryption key, and a step (f) of generating the encrypted processing model by encrypting the processing model using the encryption key, wherein in step (d), the inference system is generated which includes the encrypted processing model, the decryption key, and the configuration information.

[0017] (9) The learning method described in (7) above further comprises the steps of: (g) storing, in a storage area inaccessible to the user, an encryption key for encrypting the processing model generated in step (b) and a decryption key for decrypting the encrypted processing model encrypted with the encryption key; (h) acquiring the encryption key and the decryption key from the storage area; and (i) generating the encrypted processing model by encrypting the processing model using the acquired encryption key, wherein in step (d), the inference system is generated, the inference system including the encrypted processing model, the decryption key, and the configuration information.

[0018] (10) The learning method described in (7) or (9) above, further comprising a step (j) of acquiring recommended pre-learning data for pre-learning of the learning in step (b), and a step (k) of pre-learning a model to be learned in step (b) using the recommended pre-learning data.

[0019] (11) The learning method described in (10) above, wherein the recommended pre-training data includes at least one of pre-training data, pre-training setting parameters, and pre-training models.

[0020] (12) A learning method described in (11) above, in which data to be pre-trained in step (k) is selected and set from at least one of the pre-training data, pre-training setting parameters, and pre-training model acquired in step (j).

[0021] (13) A learning program for causing a computer to execute the learning method described in any one of (7) to (12) above.

[0022] An inference system including a processing model consisting of binary data is generated based on a processing model generated by learning using training data and configuration information required to generate an inference system that executes inference processing on the data. This prevents information leakage of the processing model and makes it easy to set up an inference environment using the processing model.

[0023] Advantages and features provided by one or more embodiments of the present invention will be more fully understood from the following detailed description and the accompanying drawings, which are for illustrative purposes only and are not intended to limit the present invention.

[0014] Figure 1 shows a schematic configuration of a learning device.

[0015] Figure 2 shows a block diagram of the hardware configuration of a learning device.

[0016] Figure 3 shows a flowchart illustrating the operation of a learning device.

[0017] Figure 4 shows a schematic configuration of a learning device.

[0018] Figure 5 shows a flowchart illustrating the operation of a learning device.

[0019] Figure 6 shows a schematic configuration of a learning device.

[0020] Figure 7 shows a flowchart illustrating the operation of a learning device.

[0021] Figure 8 shows a schematic configuration of a learning device.

[0024] Hereinafter, a learning device, a learning method, and a learning program according to embodiments of the present invention will be described with reference to the accompanying drawings. However, the scope of the present invention is not limited to the disclosed embodiments. In the description of the drawings, the same elements are designated by the same reference numerals, and duplicated explanations will be omitted. Furthermore, the dimensional proportions in the drawings are exaggerated for the sake of explanation and may differ from the actual proportions.

[0025] First Embodiment Fig. 1 is a diagram showing a schematic configuration of a learning device 100 according to the first embodiment. Fig. 2 is a block diagram showing the hardware configuration of the learning device 100.

[0026] 2, learning device 100 includes a control unit 110, a memory unit 120, a display unit 130, an input unit 140, and a communication unit 150. These components are connected to each other via a bus 160. Learning device 100 may be configured by a computer.

[0027] The control unit 110 is configured by a CPU (Central Processing Unit) and controls and performs calculations on each part of the learning device 100 according to a program. The function of the control unit 110 will be described later with reference to FIG.

[0028] The storage unit 120 may be configured with RAM (Random Access Memory), ROM (Read Only Memory), and flash memory. The RAM temporarily stores programs and data as a working area for the control unit 110. The ROM stores various programs and data in advance. The flash memory stores various programs and data, including the operating system.

[0029] The display unit 130 is, for example, a liquid crystal display, and displays various information.

[0030] The input unit 140 is configured with, for example, a touch panel and various keys, and is used for various operations and inputs.

[0031] The communication unit 150 is an interface for communicating with external devices. Network interfaces conforming to standards such as Ethernet (registered trademark), SATA, PCI Express, USB, and IEEE 1394 may be used for communication. Additionally, various local connection interfaces such as wireless communication interfaces such as Bluetooth (registered trademark) and IEEE 802.11 may also be used for communication.

[0032] The functions of the control unit 110 will now be described. As shown in Fig. 1, the control unit 11 executes a learning program to function as an input unit 111, a learning unit 112, an inference system generation unit 113, and a configuration information acquisition unit 114. That is, the control unit 11 configures the input unit 111, the learning unit 112, the inference system generation unit 113, and the configuration information acquisition unit 114 by executing the program.

[0033] The input unit 111 functions as a training data receiving unit. The input unit 111 receives input of training data. The training data may be input by a user. The user includes a user of the learning device. The user may further include a user of the inference system. The training data is data used to train the model. The training data is, for example, image data. Learning includes supervised learning and unsupervised learning. In the case of supervised learning, the training data is, for example, image data and correct labels. The model is, for example, a neural network model.

[0034] The learning unit 112 generates a processing model by learning using the learning data. Specifically, the learning unit 112 generates a processing model by learning the model using the learning data.

[0035] The inference system generation unit 113 generates an inference system including a processing model consisting of binary data based on the processing model generated by the learning unit 112 and the system configuration information of the inference system acquired by the configuration information acquisition unit 114. Hereinafter, the system configuration information of the inference system will also be simply referred to as "configuration information." The inference system is a system that performs inference processing on data to be inferred. The data to be inferred by the inference system generation unit 113 is, for example, image data. The inference processing includes processing to classify image data into one of multiple classes and processing to estimate articulation points from image data. The configuration information is information necessary for generating the inference system. The configuration information includes specifications of the computer terminal on which the inference system is executed, such as information about the CPU and GPU, and memory information. The configuration information includes configuration information of the inference system. The configuration information of the inference system includes input / output information, target language, objective information, function information, algorithms, and exception handling of the inference system.

[0036] The inference system generation unit 113 generates an inference system including a program for constructing an inference environment in a computer that constructs the inference environment, using a known method, based on the processing model and configuration information.

[0037] The inference system may be an executable file. That is, by executing the executable file on a computer terminal or the like on which the inference system is executed, an inference environment may be constructed on the computer terminal or the like. Specifically, by executing the executable file, an inference environment may be constructed that accepts input of data to be subjected to inference processing and outputs inference results according to input / output information specified in the configuration information.

[0038] The inference system may include a processing model consisting of binary data, but may not include all or part of the configuration information. In this case, all or part of the configuration information not included in the inference system may be stored in a specified storage device. When the inference system is executed, the inference system obtains all or part of the configuration information from the specified storage device, and an inference system that builds an inference environment is generated. In this case, the inference system may include a processing model consisting of binary data and information on the address of the specified storage device.

[0039] On the other hand, the inference system may include a processing model consisting of binary data and all of the configuration information acquired by the configuration information acquisition unit 114. In this case, the inference system constructs an inference environment using the processing model and configuration information included in the inference system.

[0040] The inference system generation unit 113 outputs the generated inference system. At least the processing model included in the inference system is made up of binary data, so that information about the processing model included as information in the inference system can be prevented from being leaked to users.

[0041] The configuration information acquisition unit 114 acquires the configuration information. The configuration information acquisition unit 114 can acquire the configuration information input to the input unit 140 by the user.

[0042] 3 is a flowchart showing the operation of the learning device 100. This flowchart can be executed by the control unit 110 in accordance with a program stored in the storage unit 120.

[0043] The control unit 110 accepts input of learning data by a user of the inference system (S101).

[0044] The control unit 110 generates a processing model by learning the model using the learning data (S102).

[0045] The control unit 110 acquires the configuration information input by the user (S103).

[0046] The control unit 110 generates an inference system including a processing model made up of binary data based on the processing model and the configuration information (S104).

[0047] Second Embodiment A second embodiment will be described. The present embodiment differs from the first embodiment in the following respects. In this embodiment, an encryption key for encrypting a generated processing model and a decryption key corresponding to the encryption key are generated. The processing model is then encrypted with the encryption key, and an inference system including the encrypted processing model, the decryption key, and configuration information is generated. In all other respects, this embodiment is similar to the first embodiment, and therefore, duplicated explanations will be omitted or simplified.

[0048] FIG. 4 is a diagram showing a schematic configuration of a learning device 100 according to the second embodiment.

[0049] 4, by executing the learning program, the control unit 110 functions as an input unit 111, a learning unit 112, an inference system generation unit 113, a configuration information acquisition unit 114, a key generation unit 115, and an encryption unit 116. That is, by executing the program, the control unit 11 configures the input unit 111, the learning unit 112, the inference system generation unit 113, the configuration information acquisition unit 114, the key generation unit 115, and the encryption unit 116.

[0050] The input unit 111 accepts input of learning data.

[0051] The learning unit 112 generates a processing model by learning using the learning data.

[0052] The configuration information acquisition unit 114 acquires configuration information.

[0053] The key generation unit 115 generates an encryption key for encrypting the generated processing model and a decryption key for decrypting the encrypted processing model, which is a processing model encrypted with the encryption key. Hereinafter, a processing model encrypted with an encryption key will also be simply referred to as an "encrypted processing model." The encryption key and the decryption key may be different keys or may be the same key.

[0054] The encryption unit 116 generates an encrypted processing model by encrypting the processing model using an encryption key.

[0055] The inference system generation unit 113 generates an inference system including a processing model made of binary data based on the encryption processing model and the configuration information. The generated inference system may include the encryption processing model made of binary data, a decryption key, and the configuration information.

[0056] The inference system may be an executable file. That is, by executing the executable file on a computer terminal or the like on which the inference system is executed, an inference environment may be constructed on the computer terminal or the like. Specifically, by executing the executable file, an encrypted processing model is decrypted using a decryption key to generate a processing model. Then, an inference environment may be constructed that accepts input of data to be subjected to inference processing, and outputs an inference result according to input / output information specified in the configuration information, using the processing model and configuration information.

[0057] The inference system generation unit 113 outputs the generated inference system. Since the processing model included in the inference system is made up of binary data, it is possible to prevent information about the processing model included as information in the inference system from being leaked to users of the inference system. Furthermore, the processing model included in the inference system is encrypted as an encrypted processing model. This further improves the effectiveness of preventing information about the processing model from being leaked to users.

[0058] 5 is a flowchart showing the operation of the learning device 100. This flowchart can be executed by the control unit 110 in accordance with a program stored in the storage unit 120.

[0059] The control unit 110 accepts input of learning data by a user of the inference system (S201).

[0060] The control unit 110 generates a processing model by learning the model using the learning data (S202).

[0061] The control unit 110 acquires the configuration information input by the user (S203).

[0062] The control unit 110 generates an encryption key for encrypting the processing model and a decryption key for decrypting the encrypted processing model (S204).

[0063] The control unit 110 generates an encrypted processing model by encrypting the processing model using the encryption key (S205).

[0064] Based on the encryption processing model and the configuration information, the control unit 110 generates an inference system including an encryption processing model made up of binary data, a decryption key, and configuration information (S206).

[0065] (Third Embodiment) A third embodiment will be described. The present embodiment differs from the second embodiment in the following respects. In the second embodiment, the control unit 110 generates the encryption key and the decryption key. On the other hand, in this embodiment, the encryption key and the decryption key are stored in a storage area that is inaccessible to the user, and the control unit 110 acquires the encryption key and the decryption key from this storage area. In other respects, this embodiment is similar to the first embodiment, and therefore, redundant explanations will be omitted or simplified.

[0066] FIG. 6 is a diagram showing a schematic configuration of a learning device 100 according to the third embodiment.

[0067] 6, by executing the learning program, the control unit 110 functions as an input unit 111, a learning unit 112, an inference system generation unit 113, a configuration information acquisition unit 114, a key input unit 117, and an encryption unit 116. That is, by executing the program, the control unit 11 configures the input unit 111, the learning unit 112, the inference system generation unit 113, the configuration information acquisition unit 114, the key input unit 117, and the encryption unit 116.

[0068] The input unit 111 accepts input of learning data.

[0069] The learning unit 112 generates a processing model by learning using the learning data.

[0070] The configuration information acquisition unit 114 acquires configuration information.

[0071] The key input unit 117 constitutes a key acquisition unit and acquires an encryption key and a decryption key stored in a storage area that is inaccessible to users. The encryption key is a key for encrypting a processing model, and the decryption key is a key for decrypting an encrypted processing model, which is a processing model encrypted with the encryption key. The encryption key and the decryption key may be a common key. In this case, a common key, which is the encryption key and the decryption key, may be stored in the storage area. The following description will be given taking as an example a case where a common key is stored in the storage area.

[0072] The encryption key and decryption key can be stored in advance in the storage area by the control unit 110 in accordance with instructions from an administrator or the like other than the user, input through the input unit 140. The control unit 110 constitutes a storage control unit.

[0073] The encryption unit 116 generates an encrypted processing model by encrypting the processing model using the common key.

[0074] The inference system generation unit 113 generates an inference system including a processing model made of binary data based on the encryption processing model, the configuration information, and the common key. The generated inference system may include the encryption processing model made of binary data, the decryption key, and the configuration information.

[0075] The inference system may be an executable file. That is, by executing the executable file on a computer terminal or the like on which the inference system is executed, an inference environment may be constructed on the computer terminal or the like. Specifically, by executing the executable file, an encrypted processing model is decrypted using a common key to generate a processing model. Then, an inference environment may be constructed that accepts input of data to be subjected to inference processing, and outputs an inference result according to input / output information specified in the configuration information, using the processing model and configuration information.

[0076] The inference system generation unit 113 outputs the generated inference system. Since the processing model included in the inference system is made up of binary data, it is possible to prevent information about the processing model included as information in the inference system from being leaked to users of the inference system. Furthermore, the processing model included in the inference system is encrypted as an encrypted processing model. This further improves the effectiveness of preventing information about the processing model from being leaked to users of the inference system.

[0077] 7 is a flowchart showing the operation of the learning device 100. This flowchart can be executed by the control unit 110 in accordance with a program stored in the storage unit 120.

[0078] The control unit 110 accepts input of learning data by a user of the inference system (S301).

[0079] The control unit 110 generates a processing model by learning the model using the learning data (S302).

[0080] The control unit 110 acquires the configuration information input by the user (S303).

[0081] The control unit 110 acquires an encryption key for encrypting the processing model and a decryption key for decrypting the encrypted processing model as one common key (S304).

[0082] The control unit 110 generates an encrypted processing model by encrypting the processing model using the common key (S305).

[0083] Based on the encryption processing model, the configuration information, and the common key, the control unit 110 generates an inference system including the encryption processing model made up of binary data, the common key, and the configuration information (S306).

[0084] (Fourth Embodiment) A fourth embodiment will be described. The present embodiment differs from the first embodiment in the following respects. In this embodiment, recommended pre-learning data for pre-learning by the learning unit 112 is acquired, and the learning unit 112 performs pre-learning using the recommended pre-learning data. In other respects, this embodiment is similar to the first embodiment, and therefore, overlapping descriptions will be omitted or simplified.

[0085] FIG. 8 is a diagram showing a schematic configuration of a learning device 100 according to the fourth embodiment.

[0086] 8, by executing the learning program, the control unit 110 functions as an input unit 111, a learning unit 112, an inference system generation unit 113, a configuration information acquisition unit 114, a recommended pre-learning data input unit 118, and a learning setting unit 119. That is, by executing the program, the control unit 11 configures the input unit 111, the learning unit 112, the inference system generation unit 113, the configuration information acquisition unit 114, the recommended pre-learning data input unit 118, and the learning setting unit 119.

[0087] The recommended pre-training data input unit 118 acquires recommended pre-training data from the cloud server 200. The recommended pre-training data input unit 118 constitutes a recommended pre-training data acquisition unit.

[0088] The learning setting unit 119 sets the recommended pre-learning data as learning data for pre-learning by the learning unit 112. The learning setting unit 119 constitutes a learning control unit. Pre-learning is learning by the learning unit 112 that is performed before learning using the learning data received by the input unit 111. The recommended pre-learning data may include at least one of pre-learning data, pre-learning setting parameters, and a pre-learning model. The pre-learning data is, for example, image data used for pre-learning. The pre-learning setting parameters are setting parameters used for pre-learning and necessary for learning. The pre-learning setting parameters include hyperparameters such as a learning rate. The pre-learning model is a model suitable for inference processing. By performing pre-learning and post-pre-learning learning on a model suitable for inference processing, the estimation accuracy in the inference processing can be improved.

[0089] The learning setting unit 119 may select and set data to be pre-trained by the learning unit from at least one of the pre-training data, pre-training setting parameters, and pre-training models acquired by the recommended pre-training data input unit 118. The learning setting unit 119 may set data to be pre-trained by the learning unit from at least one of the pre-training data, pre-training setting parameters, and pre-training models in accordance with instructions from an administrator other than the user input to the input unit 140.

[0090] The input unit 111 accepts input of learning data.

[0091] The learning unit 112 generates a processing model by performing pre-learning using the recommended pre-learning data, and then further learns the processing model using the training data received by the input unit 111.

[0092] The configuration information acquisition unit 114 acquires configuration information.

[0093] The inference system generation unit 113 generates an inference system including a processing model made of binary data based on the processing model and the configuration information. The generated inference system may include a processing model made of binary data and the configuration information.

[0094] The inference system may be an executable file. That is, by executing the executable file on a computer terminal or the like on which the inference system is executed, an inference environment may be constructed on the computer terminal or the like. Specifically, by executing the executable file, an inference environment may be constructed that accepts input of data to be subjected to inference processing and outputs inference results according to input / output information specified in the configuration information.

[0095] The inference system generation unit 113 outputs the generated inference system. Because the processing model included in the inference system is made up of binary data, it is possible to prevent information about the processing model included as information in the inference system from being leaked to the user of the inference system. Furthermore, the recommended pre-training data is acquired by the control unit 110, and the user cannot be involved in acquiring the recommended pre-training data. This makes it possible to prevent learning know-how included in the recommended pre-training data from being leaked to the user.

[0096] This embodiment has the following advantages.

[0097] An inference system including a processing model consisting of binary data is generated based on a processing model generated by learning using training data and configuration information required to generate an inference system that executes inference processing on the data. This prevents information leakage of the processing model and makes it easy to set up an inference environment using the processing model.

[0098] In addition, an encryption key for encrypting the generated processing model and a decryption key for decrypting the encrypted processing model encrypted with the encryption key are generated. An encrypted processing model is generated by encrypting the processing model using the generated encryption key. Then, an inference system including the encrypted processing model, the decryption key, and configuration information is generated. This further improves the effectiveness of preventing information about the processing model from being leaked to users. The complexity of generating keys in advance can be avoided.

[0099] Furthermore, an encryption key for encrypting the generated processing model and a decryption key for decrypting the encrypted processing model encrypted with the encryption key are stored in a storage area that is inaccessible to users. The encryption key and decryption key are obtained from the storage area. An encrypted processing model is generated by encrypting the processing model using the obtained encryption key. Then, an inference system including the encrypted processing model, the decryption key, and configuration information is generated. This further improves the effectiveness of preventing information about the processing model from being leaked to users. Furthermore, even if the key is leaked, it can be easily changed, preventing the leakage of technology from spreading.

[0100] In addition, recommended pre-training data for pre-training of the learning unit is obtained, and the learning unit is caused to pre-train using the recommended pre-training data, thereby preventing the leakage of learning know-how contained in the recommended pre-training data to users.

[0101] Furthermore, the recommended pre-training data includes at least one of pre-training data, pre-training setting parameters, and pre-training models, which flexibly and effectively prevents learning know-how from leaking to users.

[0102] Furthermore, the data to be pre-trained by the learning unit is selected and set from at least one of the acquired pre-training data, pre-training setting parameters, and pre-training model, which makes it possible to simply and effectively prevent learning know-how from being leaked to users.

[0103] The present invention is not limited to the above-described embodiments.

[0104] For example, the learning data may be text data or the like.

[0105] In addition, in the embodiments, part or all of the processing executed by the program may be replaced with hardware such as a circuit.

[0106] While embodiments of the present invention have been described and illustrated in detail, the disclosed embodiments are made for purposes of illustration and example only and are not intended to be limiting, and the scope of the present invention should be construed by the language of the appended claims.

[0107] This application is based on a Japanese patent application (Patent Application No. 2024-086891) filed on May 29, 2024, the disclosure of which is incorporated herein by reference in its entirety.

[0108] 100 Learning device, 110 Control unit, 111 Input unit, 112 Learning unit, 113 Inference system generation unit, 114 Configuration information acquisition unit, 115 Key generation unit, 116 Encryption unit, 117 Key input unit, 118 Recommended pre-learning data input unit, 119 Learning setting unit, 120 Memory unit, 130 Display unit, 140 Input unit, 150 Communication unit.

Claims

1. A learning device having a learning data accepting unit that accepts input of learning data; a learning unit that generates a processing model by learning using the learning data; a configuration information acquisition unit that acquires configuration information necessary to generate an inference system that performs inference processing on data; and an inference system generation unit that generates the inference system including the processing model consisting of binary data based on the processing model generated by the learning unit and the configuration information acquired by the configuration information acquisition unit.

2. The learning device described in claim 1 further comprises: a key generation unit that generates an encryption key for encrypting the processing model generated by the learning unit and a decryption key for decrypting the encrypted processing model encrypted with the encryption key; and an encryption unit that generates the encrypted processing model by encrypting the processing model using the encryption key, wherein the inference system generation unit generates the inference system including the encrypted processing model, the decryption key, and the configuration information.

3. The learning device described in claim 1 further comprises: a storage control unit that stores, in a storage area that is not accessible to users, an encryption key for encrypting the processing model generated by the learning unit, and a decryption key for decrypting the encrypted processing model encrypted with the encryption key; a key acquisition unit that acquires the encryption key and the decryption key from the storage area; and an encryption unit that generates the encrypted processing model by encrypting the processing model using the acquired encryption key, wherein the inference system generation unit generates the inference system including the encrypted processing model, the decryption key, and the configuration information.

4. A learning device as described in claim 1 or 3, further comprising: a recommended pre-learning data acquisition unit that acquires recommended pre-learning data for pre-learning of the learning unit; and a learning control unit that performs pre-learning of the learning unit using the recommended pre-learning data.

5. The learning device according to claim 4, wherein the recommended pre-training data includes at least one of pre-training data, pre-training setting parameters, and pre-training models.

6. The learning device described in claim 5, wherein the learning control unit selects and sets data to be pre-trained by the learning unit from at least one of pre-training data, pre-training setting parameters, and pre-training models acquired by the recommended pre-training data acquisition unit.

7. A method for causing a learning device to execute the method, comprising: (a) a step of accepting input of learning data; (b) a step of generating a processing model by learning using the learning data; (c) a step of acquiring configuration information necessary for generating an inference system that performs inference processing on data; and (d) a step of generating the inference system including the processing model consisting of binary data based on the processing model generated in step (b) and the configuration information acquired in step (c).

8. The learning method described in claim 7, further comprising: a step (e) of generating an encryption key for encrypting the processing model generated in step (b) and a decryption key for decrypting the encrypted processing model encrypted with the encryption key; and a step (f) of generating the encrypted processing model by encrypting the processing model using the encryption key, wherein in step (d), the inference system including the encrypted processing model, the decryption key, and the configuration information is generated.

9. The learning method described in claim 7, further comprising the steps of: (g) storing, in a storage area inaccessible to users, an encryption key for encrypting the processing model generated in step (b), and a decryption key for decrypting the encrypted processing model encrypted with the encryption key; (h) acquiring the encryption key and the decryption key from the storage area; and (i) generating the encrypted processing model by encrypting the processing model using the acquired encryption key, wherein in step (d), the inference system is generated which includes the encrypted processing model, the decryption key, and the configuration information.

10. The learning method described in claim 7, further comprising: a step (j) of acquiring recommended pre-training data for pre-learning of the learning in step (b); and a step (k) of pre-training a model to be trained in step (b) using the recommended pre-training data.

11. The learning method according to claim 10, wherein the recommended pre-training data includes at least one of pre-training data, pre-training setting parameters, and pre-training models.

12. The learning method described in claim 11, wherein data to be pre-trained in step (k) is selected and set from at least one of the pre-training data, pre-training setting parameters, and pre-training model acquired in step (j).

13. A learning program for causing a computer to execute the learning method according to any one of claims 7 to 12.

Images