Inference system, information processing system, inference method, and program

Abstract

The inference system of the present disclosure comprises: a parameter acquisition means for acquiring parameters for a plurality of models which are trained by a plurality of learning participant servers, respectively, and which are for inferring a particular event; an integration means for integrating the plurality of acquired parameters; an analysis data acquisition means for acquiring, from an inference user, analysis data for inferring a particular event; an inference means for executing inference on the basis of an integrated model, into which the plurality of parameters have been integrated, and by using the analysis data; and an output means for outputting the results of the inference.

Description

Inference system, information processing system, inference method, and recording medium

[0001] The present disclosure relates to an inference system, an information processing system, an inference method, and a recording medium.

[0002] There are efforts underway to share AI models that use artificial intelligence (AI) trained by individual companies and other organizations.

[0003] For example, Patent Literature 1 discloses a method for executing an integration process for integrating machine learning results set in a common part for each trained model. Patent Literature 2 also discloses a method for grouping users who did not participate in federated learning using the same criteria as users who participated in the federated learning, and distributing group models for each group to the electronic devices of users belonging to the group.

[0004] JP 2020-115311 A JP 2021-197181 A

[0005] However, in the invention described in Patent Document 2, a global model (integrated model) created by federated learning is directly passed to an inference user, who is a user, and can be used locally. In this case, there are issues such as the risk of the original data of the integrated model being leaked and the possibility that the AI ​​model may be used permanently without any restrictions on its use.

[0006] An example of an objective of the present disclosure is to provide a system that can manage the use of an integrated model while preventing the original data of the integrated model from leaking to inference users.

[0007] An inference system in one aspect of the present disclosure includes a parameter acquisition means for acquiring parameters of a plurality of models for inferring a specific event, which are learned in each of a plurality of learning participant servers; an integration means for integrating the acquired plurality of parameters; an analysis data acquisition means for acquiring analysis data for inferring a specific event from an inference user; an inference means for performing inference using the analysis data based on the integrated model integrating the plurality of parameters; and an output means for outputting the inference results.

[0008] An inference system in one aspect of the present disclosure includes a parameter acquisition means for acquiring analytical data for inferring a specific event, an inference means for performing inference regarding the specific event using the analytical data based on an integrated model in which parameters of multiple models are integrated through federated learning using secure computation, and an output means for outputting the inference results.

[0009] An information processing system in one aspect of the present disclosure is an information processing system having a plurality of learning participant servers and the inference system described above, wherein each learning participant server is equipped with a model storage means for storing a trained model for inferring about a specific event, an input / output means for inputting parameters of the stored model that have been updated by federated learning using secure computation in a confidential format, a restoration unit for restoring the input parameters, and an inference means for applying the restored parameters to the stored model to update the model and make inference about the specific event.

[0010] An inference method in one aspect of the present disclosure is learned in each of multiple learning participant servers, and for multiple models for inferring a specific event, parameters of the multiple models are acquired, the acquired multiple parameters are integrated, analytical data for inferring the specific event is acquired from an inference user, inference is performed using the analytical data based on the integrated model in which the multiple parameters are integrated, and the inferred inference results are output.

[0011] In one aspect of the present disclosure, a recording medium stores a program that causes a computer to perform the following steps: acquire parameters of multiple models for inferring a specific event, which are learned in each of multiple learning participant servers; integrate the acquired parameters; acquire analytical data for inferring a specific event from an inference user; perform inference using the analytical data based on the integrated model that integrates the multiple parameters; and output the inferred inference results.

[0012] One example of the effect of the present disclosure is that it can provide a system that can manage the use of an integrated model while preventing the original data of the integrated model from leaking to inference users.

[0013] Fig. 1 is a block diagram showing the configuration of an information processing system in a first embodiment. Fig. 2 is a diagram showing the hardware configuration in which an inference system in the first embodiment is realized by a computer device and its peripheral devices. Fig. 3 is a flowchart showing the operation of inference processing in the first embodiment. Fig. 4 is a block diagram showing the configuration of an information processing system in a modified example of the first embodiment. Fig. 5 is a flowchart showing the operation of information processing in a modified example of the first embodiment.

[0014] Next, an embodiment will be described in detail with reference to the drawings.

[0015] [First Embodiment] Fig. 1 is a block diagram showing the configuration of an information processing system 10 according to the first embodiment. The information processing system 10 according to the first embodiment is a system for performing inference using analytical data acquired from inference users, using an integrated model that integrates parameters of multiple models for inferring specific phenomena held by each learning participant. Examples of learning participants include organizations such as large companies and corporations that hold a large amount of training data for generating trained models. Examples of inference users include small organizations that do not hold a large amount of training data and wish to use inference using the integrated model.

[0016] Referring to FIG. 1, an information processing system 10 includes an inference system 100, multiple learning participant servers 200 (200a, 200b), and an inference user server 300 owned by an inference user. The inference system 100 inputs analytical data acquired from the inference users into an integrated model, and outputs inference results related to events. A specific event is, for example, something that can be expressed by a model (mathematical formula) of any format. However, there are no limitations on the events that the inference system 100 infers.

[0017] The inference system 100 includes a parameter acquisition unit 101, an integration unit 102, an analysis data acquisition unit 103, an inference unit 104, and an output unit 105. Each of the learning participant servers 200 includes a model generation unit 201 (201a, 201b) that generates a model for inferring a specific event, and an input / output unit 202 (202a, 202b) that inputs and outputs parameters to and from the inference system 100. In this embodiment, the multiple learning participant servers 200 are located in two locations, but this is not limited to this. The multiple learning participant servers 200 are provided in the same number as the number of organizations participating in the learning. The inference system 100, which is an essential component of this embodiment, will be described in detail below.

[0018] 2 is a diagram showing an example of a hardware configuration in which the inference system 100 according to the first embodiment of the present disclosure is realized by a computer device 500 including a processor. As shown in FIG. 2, the inference system 100 includes a CPU (Central Processing Unit) 501, memories such as a ROM (Read Only Memory) 502 and a RAM (Random Access Memory) 503, a storage device 505 such as a hard disk for storing a program 504, a communication I / F (Interface) 508 for network connection, and an input / output interface 511 for inputting and outputting data. In the first embodiment, parameter information received from each learning participant server 200 is input to the inference system 100 via the communication I / F 508.

[0019] The CPU 501 runs an operating system to control the entire inference system 100 according to the first embodiment of the present invention. The CPU 501 also reads programs and data into memory from a recording medium 506 attached to, for example, a drive device 507. The CPU 501 also functions as the parameter acquisition unit 101, integration unit 102, analysis data acquisition unit 103, inference unit 104, and output unit 105 according to the first embodiment, or as part of these units, and executes processing or commands in the flowchart shown in FIG. 3, which will be described later, based on the program.

[0020] The recording medium 506 is, for example, an optical disk, a flexible disk, a magneto-optical disk, an external hard disk, or a semiconductor memory. A part of the recording medium in the storage device is a non-volatile storage device, and the program is recorded therein. The program may also be downloaded from an external computer (not shown) connected to a communication network.

[0021] The input device 509 is realized by, for example, a mouse, a keyboard, built-in key buttons, etc., and is used for input operations. The input device 509 is not limited to a mouse, a keyboard, or built-in key buttons, and may be, for example, a touch panel. The output device 510 is realized by, for example, a display, and is used to check output.

[0022] As described above, the first embodiment shown in FIG. 1 is realized by the computer hardware shown in FIG. 2. However, the means for realizing each part of the inference system 100 in FIG. 1 is not limited to the configuration described above. Furthermore, the inference system 100 may be realized by a single physically coupled device, or by two or more physically separate devices connected by wire or wirelessly. For example, the input device 509 and the output device 510 may be connected to the computer device 500 via a network. Furthermore, the inference system 100 in the first embodiment shown in FIG. 1 can also be configured using cloud computing or the like.

[0023] The parameter acquisition unit 101 acquires trained model parameters in each of the multiple learning participant servers 200, triggered, for example, by an operation for integrating parameters by a service provider that provides an inference service. The model is a model trained by machine learning in each learning participant to output, for example, an inference result regarding a specific event. The machine learning model includes, but is not limited to, a decision tree model, a linear regression model, a logistic regression model, a neural network model, etc.

[0024] The integrating unit 102 integrates the parameters of each model. A known method can be used as the parameter integration method. For example, when integrating, the weight of the parameters corresponding to each model can be changed depending on the characteristics of each model. The integrating unit 102 applies the parameters obtained in this way to the model and stores them in the storage device 505.

[0025] The parameter acquisition unit 101 may acquire the parameters in a confidential format from each learning participant server 200. In this case, the integration unit 102 integrates multiple confidential parameters using secure computation. In this embodiment, integrating multiple confidential parameters using secure computation means that the inference system 100 performs machine learning in a state where it is distributed to each learning participant server 200 (federated learning), and integrates the parameters of the trained model using secure computation.

[0026] Methods of secure computation include special encryption methods that support specific processes, such as homomorphic encryption, a trusted execution environment that processes in an isolated state on hardware, or multi-party computation that processes secrets across multiple servers (secret sharing computation).

[0027] The following is an example of a specific method for secure computation using multi-party computation. For example, concealed data a, a parameter acquired from an arbitrary learning participant server, is secretly shared into shares x1, y1, ..., and x1, y1, ... are each transmitted by an administrator to different servers. Furthermore, concealed data b, a parameter acquired from another learning participant server, is secretly shared into shares x2, y2, ..., and x2, y2, ... are each transmitted by an administrator to different servers. Next, the concealed data a and the concealed data b communicate with each other while remaining secretly shared, and computation proceeds. Finally, the output shares u, v, ..., which are the computation results of each server, are collected and restored, resulting in the computation result F(a, b). This computation result is a parameter obtained by integrating the parameters of each model. For this reason, when multi-party computation is used as a secure computation method, the integrating unit 102 includes multiple servers. Multi-party computation eliminates the need for encryption key management or an isolated environment, and enables faster computation. The integration unit 102 restores the model parameters thus obtained, and stores the model to which the restored parameters have been applied in the storage device 505 .

[0028] The analytical data acquisition unit 103 is a means for acquiring analytical data relating to a specific event from an inference user. The analytical data is, for example, data to be input into an integrated model for inferring a specific event. The analytical data acquisition unit 103 acquires analytical data from the inference user, for example, by storing the analytical data in a predetermined server prepared for executing inference. The analytical data acquisition unit 103 outputs the acquired analytical data to the inference unit 104.

[0029] The analytical data acquisition unit 103 may acquire analytical data only within a predetermined period. The predetermined period is, for example, a period during which the inference user uses the inference service established between the inference user and the service provider. In this case, the inference user may be configured to be able to access the server for storing analytical data only within the predetermined period.

[0030] The inference unit 104 is a means for executing inference using analytical data based on an integrated model that integrates multiple parameters. The inference unit 104 executes inference by inputting the analytical data into the integrated model stored in the storage device 505. The inference unit 104 outputs the inference results to the output unit 105.

[0031] The output unit 105 is a means for outputting the inference result. The output unit 105 stores the inference result in a server accessible by the inference user. The output unit 105 may also transmit the inference result to the inference user server 300.

[0032] The analytical data acquisition unit 103 may acquire analytical data in a concealed format from the inference user. In this case, the inference unit 104 performs inference by performing secure computation on the analytical data while keeping it concealed. As a method of secure computation, the same secure computation as used in the case of integrating the model parameters described above can be used. Then, the output unit 105 outputs the inference result to the inference user in a concealed format. In this case, the inference user needs to restore the concealed inference result.

[0033] The operation of the inference system 100 configured as above will be described with reference to the flowchart of FIG.

[0034] 3 is a flowchart showing an outline of the operation of the inference system 100 in the first embodiment. The processing according to this flowchart may be executed based on program control by the processor described above. Furthermore, the series of processes according to this flowchart do not have to be performed consecutively. For example, the parameter integration in steps S101 to S104 in FIG. 3 and the inference in steps S105 to S110 may be performed at different times. Furthermore, the inference in steps S105 to S110 may be performed multiple times, in which case the inference may be performed by different inference users.

[0035] 3 , first, the parameter acquisition unit 101 acquires parameters of a trained model for inferring a specific event from each learning participant server 200 (step S101). If the multiple parameters acquired by the parameter acquisition unit 101 are in a concealed format (S102; YES), the integration unit 102 integrates the multiple concealed model parameters by secure computation (step S103). On the other hand, if the multiple parameters acquired by the parameter acquisition unit 101 are not in a concealed format (S102; NO), the integration unit 102 integrates the multiple parameters without using secure computation (step S104).

[0036] Next, the analytical data acquisition unit 103 acquires analytical data for inferring a specific event from the inference user (step S105). If the analytical data acquisition unit 103 acquires analytical data in a concealed format from the inference user (S106; YES), the inference unit 104 performs inference by performing secure computation on the analytical data while keeping it concealed, based on an integrated model that integrates multiple parameters. Next, the output unit 105 outputs the inference result by the inference unit 104 in a concealed format (step S108). On the other hand, if the analytical data acquisition unit 103 acquires analytical data that is not concealed from the inference user (S106; NO), the inference unit 104 performs inference using the analytical data based on the integrated model (step S109), and then the output unit 105 outputs the inference result by the inference unit 104 (step S110). This completes the inference operation of the inference system 100.

[0037] In the inference system 100, the inference unit 104 performs inference using the analysis data of the inference user based on an integrated model that integrates multiple parameters. Therefore, inference can be performed in the inference system 100 without providing the integrated model to the inference user. This makes it possible to provide a system that can manage the use of the integrated model while preventing the original data of the integrated model from leaking to the inference user.

[0038] Furthermore, in the inference system 100, when the multiple parameters acquired by the parameter acquisition unit 101 are in a confidential format, the integration unit 102 integrates the multiple confidential parameters using secure computation. This allows the integrated model to be used while keeping the parameters of each model confidential.

[0039] Furthermore, in the inference system 100, when the analytical data acquisition unit 103 acquires analytical data in a concealed format from an inference user, the inference unit 104 executes inference by performing secret calculations on the analytical data while keeping it concealed, based on an integrated model that integrates multiple parameters, and the output unit 105 outputs the inference result in a concealed format. This makes it possible to avoid the risk of know-how contained in the analytical data of the inference user being leaked.

[0040] Next, applications of the inference system 100 of the present disclosure will be described. The inference system 100 can be applied to various fields, such as agriculture, fisheries, construction, mobility, finance, education, medicine, healthcare, manufacturing equipment, and chemistry. Applications of the inference system 100 of the present disclosure in each field will be described below, but these applications are merely examples of the present disclosure. Each learning participant is grouped, for example, by the application of the model they generated, and participates in federated learning if the application matches the event inferred by the inference system 100.

[0041] In the agricultural field, for example, learning participants include large agricultural corporations and smart agriculture operators utilizing ICT (Information and Communication Technology), while inference users include small and medium-sized agricultural corporations. Examples of data used for inference, such as training data and analysis data, include room temperature, humidity, weather, carbon dioxide levels, crop nutrient data, or soil data, and are used for growth prediction, growth rate / quantity control, or disease prediction. The model used for inference in the inference system 100 is a model that inputs, for example, room temperature, humidity, weather, carbon dioxide levels, crop nutrient data, or soil data, and outputs growth prediction, growth rate / quantity control, or disease prediction.

[0042] In the fishing industry, for example, learning participants include large fishing corporations and smart fishing operators that utilize ICT, and inference users include small and medium-sized fishing corporations. Data used for inference includes, for example, seawater temperature, weather, fish movement, feed consumption, or fish growth data, and is used for growth prediction, growth rate control, or disease prediction. A model used for inference in the inference system 100 is a model that inputs, for example, seawater temperature, weather, fish movement, feed consumption, or fish growth data, and outputs growth prediction, growth rate control, or disease prediction.

[0043] In the construction field, for example, learning participants include major construction machinery manufacturers or major general contractors, and inference users include mid-sized general contractors or small- to medium-sized construction companies. Data used for inference includes construction machinery operation data or site environmental data, and is used, for example, to predict failures, recommend man-hour improvements, or infer environmental impacts. A model used for inference in the inference system 100 is, for example, a model that inputs construction machinery operation data or site environmental data and outputs failure predictions, recommendations for man-hour improvements, or environmental impacts.

[0044] In the mobility field, for example, learning participants include major automobile manufacturers or map development companies, and inference users include users (drivers) and transportation operators. Data used for inference includes road condition data, weather, or traffic information, and is used for inferring, for example, automated driving assistance, hazard prediction, accident prevention, and driver malfunction detection. A model used for inference in the inference system 100 is, for example, a model that inputs road condition data, traffic information, or environmental information and outputs hazard prediction.

[0045] In the financial field, for example, megabanks can be cited as learning participants, and regional banks can be cited as inference users. Data used for inference can include transaction history or customer attributes, and can be used, for example, to infer anti-money laundering or fraud detection. A model used for inference in the inference system 100 is, for example, a model that inputs the details of a financial transaction, such as a customer's remittance, and outputs whether or not the transaction constitutes fraud such as money laundering.

[0046] In the field of education, for example, learning participants include large cram schools, preparatory schools, or correspondence education providers, and inference users include individuals or small- to medium-sized cram schools. Data used for inference includes learning data indicating students' learning volume and grades, and is used, for example, to infer adaptive learning or pass rate prediction. A model used for inference in the inference system 100 is, for example, a model that inputs a subject's learning volume and grades and outputs a pass rate prediction.

[0047] In the medical field, for example, learning participants include large hospitals and university hospitals, and inference users include small and medium-sized hospitals and clinics. Data used for inference includes medical information from electronic medical records and image information such as CT (Computed Tomography) and MRI (Magnetic Resonance Imaging), and is used, for example, to support diagnosis. A model used for inference in the inference system 100 is, for example, a model that inputs at least one of medical information from a patient's electronic medical record, CT image information, and MRI image information, and outputs a diagnosis result for the patient.

[0048] In the field of manufacturing equipment, for example, learning participants include manufacturing equipment manufacturers, and inference users include equipment installation companies. Data used for inference include equipment operation data or manufacturing information, and applications include, for example, failure prediction or improving yield. A model used for inference in the inference system 100 is, for example, a model that inputs equipment operation data and outputs failure prediction.

[0049] In the field of chemistry, for example, learning participants include major chemical manufacturers, and inference users include small and medium-sized chemical manufacturers. Data used for inference includes analytical and experimental data, and is used for purposes such as improving the efficiency of material development and predicting target performance or characteristics. A model used for inference in the inference system 100 is, for example, a model that inputs information about the material to be developed and outputs either the efficiency of material development, target performance, predicted characteristics of the material, or a synthesis method.

[0050] [Variations of the First Embodiment] Next, variations of the first embodiment of the present disclosure will be described in detail with reference to the drawings. Below, descriptions of content that overlaps with the above description will be omitted to the extent that the description of this embodiment is not unclear. In the variations of the first embodiment, the information processing system 11 is used to provide each learning participant server with an updated model through federated learning using secure computation. These updated models are used, for example, by each learning participant to infer specific events. The functions of each component in each embodiment of the present disclosure can be realized not only by hardware, as with the computer device shown in FIG. 2, but also by a computer device based on program control.

[0051] <Inference System> Figure 4 is a block diagram showing the configuration of an information processing system 11 including an inference system 110 according to a modified example of the first embodiment of the present disclosure. With reference to Figure 4, the inference system 110 according to the modified example of the first embodiment, the learning participant servers 210 (210a, 210b), and the inference user server 310 will be described, focusing on differences from the information processing system 10 according to the first embodiment. The inference system 110 includes a parameter acquisition unit 111, an integration unit 112, a parameter transmission unit 113, an analysis data acquisition unit 114, an inference unit 115, and an output unit 116.

[0052] The parameter acquisition unit 111 acquires parameters of the trained model of each learning participant from the learning participant server 210 via the communication I / F 508. Next, the integration unit 112 integrates the received parameters of the concealed models using secure computation and outputs the integrated parameters of the concealed model to the parameter transmission unit 113 in a concealed format. The parameter transmission unit 113 transmits the integrated parameters to each learning participant server 210 via the input / output unit 213. Furthermore, after transmitting the parameters to the learning participant server 210, if the learning participant server 210 retrains the model and updates the parameters, the inference system 110 may acquire the updated parameters again. Note that the operations of the analysis data acquisition unit 114, the inference unit 115, and the output unit 116 are similar to the operations of the corresponding components in the first embodiment, and therefore will not be described here.

[0053] <Learning participant server> In a variation of the first embodiment, multiple learning participant servers 210 (210a, 210b) include a model generation unit 211 (211a, 211b), a concealment unit 212 (212a, 212b), an input / output unit 213 (213a, 213b), a restoration unit 214 (214a, 214b), a model storage unit 215 (215a, 215b), and a participant inference unit 216 (216a, 216b). The model storage unit 215 stores a model generated by the model generation unit 211.

[0054] The learning participant server 210 updates the model stored in the model storage unit 215 to a model that applies the parameters received from the inference system 110. Specifically, the input / output unit 213 receives the parameters in a confidential format and outputs them to the restoration unit 214. The restoration unit 214 then restores the parameters and replaces them with the parameters of the model stored in the model storage unit 215. The participant inference unit 216 then performs inference using the updated model. To improve the accuracy of the inference results of the participant inference unit 216, the learning participant server 210 may perform re-learning based on additionally acquired learning data and transmit the updated parameters to the inference system 110. The accuracy of the model can be further improved by repeating the parameter updating through learning in each learning participant server 210 and the parameter integration in the inference system 110, for example, until a predetermined condition is met. The predetermined condition is stored, for example, in the storage device 505.

[0055] The operation of the information processing system 11 configured as above will be described with reference to the flowchart of FIG.

[0056] 5 is a flowchart showing an outline of the operation of the information processing system 11 in the first embodiment. Note that the processing according to this flowchart may be executed based on program control by the processor described above.

[0057] As shown in FIG. 5 , first, in the learning participant server 210, the model generation unit 211 locally generates a model using data held by the learning participant (step S201). Next, the anonymization unit 212 anonymizes the model parameters (step S202), and the input / output unit 213 outputs the anonymized parameters to the inference system 110 (step S203). Next, the parameter acquisition unit 111 of the inference system 110 acquires the anonymized parameters (step S204). Next, the integration unit 112 integrates the multiple anonymized parameters using secure computation (step S205). Next, the parameter transmission unit 113 outputs the model parameters integrated by the integration unit 112 to each learning participant server 210 in anonymized form (step S206).

[0058] Next, each learning participant server 210 acquires the integrated parameters in a confidential format through the input / output unit 213 (step S207). Next, the restoration unit 214 restores the parameters in a confidential format (step S208). Next, the learning participant server 210 updates the model stored in the model storage unit 215 to a model that applies the restored parameters (step S209). Next, the learning participant server 210 determines whether a predetermined condition is met (step S210). If the predetermined condition is met (step S210; YES), the participant inference unit 216 performs inference using the updated model (step S211). If the predetermined condition is not met, the learning participant server 210 returns to step S201 (step S210; NO) and executes the flow again.

[0059] On the other hand, when the inference user server 310 requests inference on a specific event, it transmits analytical data in a confidential format to the inference system 100 (step S212). Next, in the inference system 100, the analytical data acquisition unit 114 acquires the analytical data in a confidential format (step S213), the inference unit 115 performs secure computation while keeping the analytical data confidential (step S214), and the output unit 116 outputs the inference result to the inference user server 310 in a confidential format (step S215).

[0060] When the inference user server 310 receives the inference result in a concealed format (step S216), it restores the inference result (step S217). With this, the information processing system 11 ends the information processing operation.

[0061] In a modification of the first embodiment of the present disclosure, the participant inference unit 216 of each learning participant server 210 performs inference on a specific event using a model to which parameters of the model integrated by the integration unit 112 are applied. This allows each learning participant server 210 to output a more accurate inference result.

[0062] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

[0063] For example, although multiple operations are described in order in the form of a flowchart, the order of description does not limit the order in which the multiple operations are performed. Therefore, when implementing each embodiment, the order of the multiple operations can be changed to the extent that it does not interfere with the content. Specifically, in the flowchart of Figure 5, after the integration unit 112 in the inference system 110 integrates parameters, the parameter transmission unit 113 transmits the integrated parameters to the learning participant server 210. However, after the integration unit 112 in the inference system 110 integrates the parameters, the analysis data acquisition unit 114 may first acquire analysis data from the inference user server 310.

[0064] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0065] (Supplementary Note 1) An inference system comprising: a parameter acquisition means for acquiring parameters of a plurality of models for inferring a specific event, the parameters being learned in each of a plurality of learning participant servers; an integration means for integrating the acquired parameters; an analysis data acquisition means for acquiring analysis data for inferring the specific event from an inference user; an inference means for performing inference using the analysis data based on the integrated model obtained by integrating the parameters; and an output means for outputting the inference results.

[0066] (Appendix 2) An inference system as described in Appendix 1, wherein the analytical data acquisition means acquires the analytical data in a confidential format from the inference user, the inference means performs inference by performing secret calculations on the analytical data while keeping it confidential, and the output means outputs the inference results in a confidential format.

[0067] (Supplementary Note 3) The inference system according to Supplementary Note 1 or Supplementary Note 2, wherein the parameter acquisition means acquires parameters of the plurality of models in a confidential format, and the integration means integrates the plurality of parameters using secure computation.

[0068] (Supplementary Note 4) An inference system according to any one of Supplementary Notes 1 to 3, wherein the analytical data acquisition means acquires the analytical data only within a predetermined period of time.

[0069] (Supplementary Note 5) An inference system comprising: analytical data acquisition means for acquiring analytical data for inferring a specific event; inference means for performing inference regarding the specific event using the analytical data based on an integrated model in which parameters of multiple models are integrated by federated learning using secure computation; and output means for outputting the inference results.

[0070] (Supplementary Note 6) The inference system according to Supplementary Note 5, wherein the inference means performs inference by performing a secure calculation on the analysis data while keeping the analysis data confidential.

[0071] (Appendix 7) An inference system described in any of Appendices 1 to 6, wherein the model is a model that inputs medical information from a patient's electronic medical record, CT image information, or MRI image information as the analysis data, and outputs a diagnosis result for the patient.

[0072] (Appendix 8) An inference system according to any one of appendices 1 to 6, wherein the model is a model that inputs road condition data, traffic information, or environmental information as the analysis data, and outputs a danger sign.

[0073] (Appendix 9) An inference system described in any of Appendices 1 to 6, wherein the model inputs the content of a customer's financial transaction as the analysis data and outputs whether or not it constitutes fraud, including money laundering.

[0074] (Appendix 10) An inference system according to any one of Appendices 1 to 6, wherein the model is a model that inputs information about a material to be developed as the analytical data and outputs one of efficiency improvements in material development, target performance, predicted properties of the material, or a synthesis method.

[0075] (Supplementary Note 11) An information processing system having a plurality of learning participant servers and an inference system according to any one of Supplements 1 to 10, wherein each of the learning participant servers comprises: a model storage means for storing a trained model for inferring about a specific event; an input / output means for inputting, in a confidential format, parameters of the stored model that have been updated by federated learning using secure computation; a restoration means for restoring the input parameters; and a participant inference means for applying the restored parameters to the stored model to update the model and make an inference about the specific event.

[0076] (Supplementary Note 12) An inference method comprising: acquiring parameters of a plurality of models for inferring a specific event, which are learned in each of a plurality of learning participant servers; integrating the acquired parameters; acquiring analytical data for inferring the specific event from an inference user; performing inference using the analytical data based on the integrated model in which the parameters are integrated; and outputting the inferred inference results.

[0077] (Appendix 13) A recording medium storing a program that causes a computer to execute the following steps: acquire parameters of a plurality of models for inferring a specific event, which are learned in each of a plurality of learning participant servers; integrate the acquired parameters; acquire analytical data for inferring the specific event from an inference user; perform inference using the analytical data based on the integrated model that integrates the parameters; and output the inference results.

[0078] 10, 11 Information processing system 100, 110 Inference system 101, 111 Parameter acquisition unit 102, 112 Integration unit 103, 114 Analysis data acquisition unit 104, 115 Inference unit 105, 116 Output unit 113 Parameter transmission unit 200, 210 Learning participant server 201, 211 Model generation unit 202, 213 Input / output unit 212 Concealment unit 214 Restoration unit 215 Model storage unit 216 Participant inference unit 300, 310 Inference user server

Claims

1. A parameter acquisition means for acquiring parameters of a plurality of models for inferring a specific event, the models being trained in each of the plurality of learning participant servers; An integration means for integrating the acquired multiple parameters; an analysis data acquisition means for acquiring analysis data for inferring the specific event from an inference user; an inference means for performing inference using the analysis data based on an integrated model obtained by integrating the plurality of parameters; An inference system comprising: an output means for outputting an inference result.

2. The analytical data acquisition means acquires the analytical data in a concealed format from the inference user, the inference means performs inference by performing a secret calculation on the analysis data while keeping the analysis data concealed; The inference system according to claim 1 , wherein the output means outputs the inference result in a concealed format.

3. The parameter acquisition means acquires parameters of the plurality of models in a concealed format; The inference system according to claim 1 , wherein the integration means integrates the plurality of parameters using secure computation.

4. The inference system according to claim 1 , wherein the analytical data acquisition means acquires the analytical data only within a predetermined period of time.

5. An analytical data acquisition means for acquiring analytical data for inferring a specific event; an inference means for executing an inference regarding the specific event using the analysis data based on an integrated model in which parameters of a plurality of models are integrated by federated learning using secure computation; An inference system comprising: an output means for outputting an inference result.

6. The inference system according to claim 5 , wherein the inference means performs inference by performing a secure calculation on the analysis data while keeping the analysis data confidential.

7. The inference system according to claim 1 , wherein the model inputs, as the analysis data, one of medical information from a patient's electronic medical record, CT image information, and MRI image information, and outputs a diagnosis result for the patient.

8. The inference system according to claim 1 , wherein the model is a model that inputs, as the analysis data, any one of road condition data, traffic information, and environmental information, and outputs a danger sign.

9. The inference system according to claim 1 , wherein the model inputs the details of a customer's financial transaction as the analysis data and outputs whether or not the transaction constitutes fraud, including money laundering.

10. The inference system of claim 1 , wherein the model inputs information about the material to be developed as the analytical data, and outputs either efficiency of material development, target performance, predicted properties of the material, or a synthesis method.

11. An information processing system having a plurality of learning participant servers and the inference system according to any one of claims 1 to 10, Each of the learning participant servers has a model storage means for storing a trained model for making inferences regarding a specific event; an input / output means for inputting, in a concealed format, parameters updated by federated learning using secure computation for the stored parameters of the model; A restoring means for restoring the input parameters; and a participant reasoning means for applying the restored parameters to the stored model to update the model and to make a reasoning regarding the particular event.

12. acquiring parameters of a plurality of models for inferring a specific event, the models being learned in each of the plurality of learning participant servers; Integrating the acquired multiple parameters; acquiring analysis data for inferring the specific event from an inference user; performing inference using the analytical data based on an integrated model that integrates the plurality of parameters; An inference method that outputs the inferred inference result.

13. acquiring parameters of a plurality of models for inferring a specific event, the models being learned in each of the plurality of learning participant servers; Integrating the acquired multiple parameters; acquiring analysis data for inferring the specific event from an inference user; performing inference using the analytical data based on an integrated model that integrates the plurality of parameters; A program that causes a computer to execute an inference and output the results of the inference.