Information processing system

Abstract

To provide an information processing system for protecting the right to a result generated by each client in associative learning.SOLUTION: An information processing system includes a distributed network for achieving a distributed ledger. The distributed ledger stores a transaction related to a result generated from machine learning by one client among a plurality of clients performing associative learning of a machine learning model. The transaction includes timing information on the result.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to the technical field of information processing systems.

Background Art

[0002] As this type of system, for example, a system has been proposed that monitors the state information of a computer device equipped with a plurality of federated learning models, and executes a learning schedule for the plurality of federated learning models in consideration of the state information and the requirements of each of the plurality of federated learning models (see Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In federated learning, the central server updates the learning model based on the products generated by each of a plurality of clients performing machine learning. Here, there is a situation where the rights regarding the above products are not always protected.

[0005] The present invention has been made in view of the above circumstances, and an object thereof is to provide an information processing system capable of protecting the rights regarding the products generated by each client in federated learning.

Means for Solving the Problems

[0006] An information processing system according to an aspect of the present invention includes a distributed network for realizing a distributed ledger, and the distributed ledger stores a transaction related to an artifact generated by performing machine learning by one of a plurality of clients that perform federated learning of a machine learning model, and the transaction includes time information related to the artifact.

Brief Description of Drawings

[0007]

Figure 1

Figure 2

Figure 3

Embodiments for Carrying Out the Invention

[0008] Embodiments related to the information processing system will be described with reference to FIGS. 1 to 3.

[0009] (Federated Learning) First, federated learning will be described with reference to FIG. 1. FIG. 1 is a block diagram showing a system 50 for executing federated learning. In FIG. 1, the system 50 includes a central server 51 and a plurality of clients 52#1, 52#2, …, 52#n. The central server 51 and the plurality of clients 52#1, 52#2, …, 52#n may be configured to be communicable with each other via a wide area network such as the Internet. In the description common to the plurality of clients 52#1, 52#2, …, 52#n, it will be referred to as "client 52" as appropriate.

[0010] The central server 51 may have a processor 511 and a memory 512. A computer program 5121 may be stored in the memory 512. The processor 511 may execute the processing that the central server 51 should perform, together with the memory 512 in which the computer program 5121 is stored (in other words, together with the memory 512 and the computer program 5121 stored in the memory 512). For example, by executing the computer program 5121, logical functional blocks for executing the processing that the central server 51 should perform may be realized within the processor 511. A learning model 5122 may be stored in the memory 512.

[0011] The client 52#1 may have a processor 521#1 and a memory 522#1. A computer program 5221#1 may be stored in the memory 522#1. The processor 521#1 may execute the processing that the client 52#1 should perform, together with the memory 522#1 in which the computer program 5221#1 is stored (in other words, together with the memory 522#1 and the computer program 5221#1 stored in the memory 522#1). For example, by executing the computer program 5221#1, logical functional blocks for executing the processing that the client 52#1 should perform may be realized within the processor 521#1. A learning model 5222#1 may be stored in the memory 522#1.

[0012] The client 52#2 may have a processor 521#2 and a memory 522#2. A computer program 5221#2 may be stored in the memory 522#2. The processor 521#2 may execute the processing that the client 52#2 should perform, together with the memory 522#2 in which the computer program 5221#2 is stored. A learning model 5222#2 may be stored in the memory 522#2.

[0013] Client 52#n may have a processor 521#n and a memory 522#n. A computer program 5221#n may be stored in the memory 522#n. The processor 521#n may execute the processing that the client 52#n should perform together with the memory 522#n in which the computer program 5221#n is stored. A learning model 5222#n may be stored in the memory 522#n.

[0014] The processor 521#1 of client 52#1 may perform machine learning of the learning model 5222#1 using learning data. The processor 521#2 of client 52#2 may perform machine learning of the learning model 5222#2 using learning data. The processor 521#n of client 52#n may perform machine learning of the learning model 5222#n using learning data.

[0015] The plurality of clients 52#1, 52#2, …, 52#n may send the products generated by performing machine learning to the central server 51. The processor 511 of the central server 51 may update the learning model 5122 based on the plurality of products. Here, the "product" may be the learning model after machine learning, may be the parameters related to the learning model after machine learning, or may be the error information for updating the learning model 5122.

[0016] The central server 51 may send the updated learning model 5122 to the plurality of clients 52#1, 52#2, …, 52#n. Each of the plurality of clients 52#1, 52#2, …, 52#n may update the learning model (for example, learning models 5222#1, 5222#2, and 5222#n) based on the updated learning model 5122. Each of the plurality of clients 52#1, 52#2, …, 52#n may perform machine learning of the learning model updated based on the updated learning model 5122 using learning data.

[0017] Furthermore, the learning model (i.e., the learning model related to federated learning) may be, for example, a learning model applicable to an automatic driving system of a vehicle. The learning model may be, for example, a learning model applicable to a navigation device. Furthermore, the learning model is not limited to a learning model applicable to at least one of the automatic driving system and the navigation device. Furthermore, the learning model 5122 may be referred to as a global model. For example, the learning models 5222#1, 5222#2, and 5222#n may be referred to as local models.

[0018] (Information Processing System 1) Next, the information processing system 1 will be described with reference to FIGS. 2 and 3. FIG. 2 is a block diagram showing the configuration of the information processing system 1. FIG. 3 is a conceptual diagram showing the concept of the distributed ledger 3122.

[0019] In FIG. 2, the information processing system 1 includes a management server 10, a database 20, and a distributed network 30. The distributed network 30 is a network for realizing a distributed ledger. In the present embodiment, a blockchain is cited as an example of the distributed ledger.

[0020] The management server 10 may have a processor 11 and a memory 12. A computer program 121 may be stored in the memory 12. The processor 11 may execute the processing that the management server 10 should perform together with the memory 12 in which the computer program 121 is stored (in other words, together with the memory 12 and the computer program 121 stored in the memory 12). For example, by executing the computer program 121, the processor 11 may realize a logical functional block for executing the processing that the management server 10 should perform within the processor 11.

[0021] The distributed network 30 has nodes 31, 32, 33, and 34. Note that the number of nodes in the distributed network 30 is not limited to "4". Node 31 may have a processor 311 and a memory 312. A computer program 3121 may be stored in the memory 312. The processor 311 may execute the processing to be performed by the node 31 together with the memory 312 in which the computer program 3121 is stored (in other words, together with the memory 312 and the computer program 3121 stored in the memory 312). For example, by executing the computer program 3121, a logical functional block for executing the processing to be performed by the node 31 may be realized in the processor 311. A distributed ledger 3122 realized by the distributed network 30 (in other words, constructed within the distributed network 30) may be stored in the memory 312 of the node 31. As shown in FIG. 3, one or more transactions Tx may be stored in the distributed ledger 3122. Note that the configurations of the nodes 32, 33, and 34 may be the same as the configuration of the node 31.

[0022] For example, among a plurality of clients 52#1, 52#2, …, 52#n, an operator of one client 52 may register, in the distributed ledger 3122, work product information regarding a work product generated by performing machine learning of a learning model by the one client 52. In this case, the operator may instruct the management server 10 to register the work product information via a terminal device (not shown). In this case, the processor 11 of the management server 10 may generate a transaction regarding the work product information.

[0023] Alternatively, one client 52 may instruct the management server 10 to register the work product information. At this time, the one client 52 may automatically (in other words, without an operator) instruct the management server 10 to register the work product information. In this case, the processor 11 of the management server 10 may generate a transaction regarding the work product information according to the instruction from the one client 52.

[0024] Alternatively, a client 52 may generate a transaction regarding deliverable information. In this case, the information processing system 1 may not include the management server 10. In this case, a client 52 may form part of the distributed network 30. For example, a client 52 may function as a node of the distributed network 30.

[0025] For example, the processor 311 of the node 31 included in the distributed network 30 may store a transaction regarding deliverable information in the distributed ledger 3122 realized by the distributed network 30.

[0026] The deliverable information may include, for example, information indicating a deliverable. The information indicating a deliverable may be, for example, identification information for identifying a deliverable, a hash value generated from a deliverable, or the deliverable itself. The deliverable information may include, for example, time information regarding a deliverable. The time information may include, for example, at least one of a learning start date and time indicating the date and time when machine learning of a learning model that is a local model started and a learning end date and time indicating the date and time when machine learning of a learning model that is a local model ended. Since a deliverable is generated by performing machine learning of a learning model that is a local model, it can be said that the learning start date and time indicates the date and time when the machine learning performed to generate the deliverable started. Similarly, it can be said that the learning end date and time indicates the date and time when the machine learning performed to generate the deliverable ended. The deliverable information may include, for example, position information indicating the position of a client 52.

[0027] Therefore, a transaction related to work product information may include information indicating a work product. As described above, the information indicating a work product may be the work product itself. In this case, a transaction related to work product information may include a learning model in which machine learning has been performed (in other words, a trained model). A transaction related to work product information may include time information related to the work product. For example, a transaction related to work product information may include at least one of a learning start date and time and a learning end date and time. Note that the learning start date and time and the learning end date and time may be referred to as time stamps. A transaction related to work product information may include position information indicating the position of a client 52. That is, a transaction related to work product information may include at least one of information indicating a work product, time information (for example, at least one of a learning start date and time and a learning end date and time), and position information.

[0028] (Technical effect) Federated learning is one of the effective means for efficiently promoting machine learning. On the other hand, for example, among a plurality of participants participating in federated learning, there is a possibility that a malicious participant may illegally use the work products of other participants. As described above, in the information processing system 1, a transaction related to work product information is stored in the distributed ledger 3122. For example, during a dispute between a malicious participant and another participant, the other participant can be expected to assert its rights based on the transaction related to the work product information stored in the distributed ledger 3122. In particular, when the transaction related to the work product information includes time information related to the work product (for example, at least one of a learning start date and time and a learning end date and time), the order relationship between another participant (that is, a legitimate right holder) and a malicious participant can be relatively easily clarified based on the transaction related to the work product information. Therefore, according to the information processing system 1, the rights related to the work products generated by each client 52 in federated learning can be protected.

[0029] In addition, the distributed ledger 3122 in which transactions related to work product information are stored clarifies the rights relationship regarding the work product. Therefore, it can be said that the rights relationship is also clear for a global model (for example, the learning model 5122) that uses a plurality of work products with a clear rights relationship. For example, it can be clarified that no illegal data is used in the construction of the global model. As a result, according to the information processing system 1, the reliability of the global model can be improved.

[0030] Various aspects of the invention derived from the embodiments described above will be described below.

[0031] An information processing system according to an aspect of the invention includes a distributed network for realizing a distributed ledger, and the distributed ledger stores a transaction related to a work product generated by performing machine learning by one of a plurality of clients that perform federated learning of a machine learning model, and the transaction includes time information related to the work product. In the above-described embodiment, "distributed network 30" corresponds to an example of a "distributed network", and "distributed ledger 3122" corresponds to an example of a "distributed ledger".

[0032] The time information may include at least one of a learning start date and time indicating the date and time when the machine learning performed to generate the work product was started and a learning end date and time indicating the date and time when the machine learning performed to generate the work product was completed. The transaction may include position information indicating the position of the one client. The transaction may include a learned model as the work product.

[0033] An information processing system according to another aspect of the invention includes a generation means for generating a transaction related to a work product generated by performing machine learning by one of a plurality of clients that perform collaborative learning of a machine learning model, and a storage means for storing the transaction in a distributed ledger. The transaction includes time information related to the work product. In the above-described embodiment, at least one of the "management server 10" and the "client 52" corresponds to an example of the "generation means", and the "node 31" corresponds to an example of the "storage means".

[0034] An information processing apparatus according to one aspect of the invention is an information processing apparatus that constitutes a part of a distributed network, and includes a storage means for storing a transaction related to a work product generated by performing machine learning by one of a plurality of clients that perform collaborative learning of a machine learning model in a distributed ledger constructed within the distributed network. The transaction includes time information related to the work product. In the above-described embodiment, the "node 31" corresponds to an example of the "information processing apparatus", and the "processor 311" corresponds to an example of the "storage means".

[0035] The present invention is not limited to the above-described embodiments, and can be appropriately modified within a range not contrary to the gist or idea of the invention that can be read from the claims and the entire specification. An information processing system involving such modifications is also included in the technical scope of the present invention.

Description of Reference Numerals

[0036] 1... Information processing system, 10... Management server, 30... Distributed network

Claims

1. comprising a distributed network for realizing a distributed ledger, wherein a transaction regarding an artifact generated by performing machine learning by one of a plurality of clients performing federated learning of a machine learning model is stored in the distributed ledger, the transaction including time information regarding the artifact, an information processing system characterized by the above.

2. the time information including at least one of a learning start date and time indicating the date and time when the machine learning performed to generate the artifact was started and a learning end date and time indicating the date and time when the machine learning performed to generate the artifact was ended, the information processing system according to claim 1, characterized by the above.

3. the transaction including location information indicating the location of the one client, the information processing system according to claim 1, characterized by the above.

4. the transaction including the learned model as the artifact, the information processing system according to claim 1, characterized by the above.

Images