JP7111343B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

Research on artificial intelligence (AI) using information networks such as neural networks is underway, and application to industrial fields such as stock trading, automatic driving, smart grids, and IoT is underway. Since an information network is usually created by a single entity (company, organization, individual, etc.), there are limits to the information that can be obtained and the processing capacity, and the profit that can be obtained as a whole is limited.

On the other hand, multi-agent reinforcement learning (MARL) is known (Non-Patent Documents 1 to 5). According to these techniques, remuneration is evenly distributed to each agent, or the amount of remuneration is determined by a trusted third party (TTP).

Non-Patent Document 1: Foerster, Jakob, Assael, Yannis, de Freitas, Nando, and Whiteson, Shimon. Learning to communicate with deep multi-agent reinforcement learning. In Advances in Neural Information Processing Systems, pp. 2137-2145, 2016.
Non-Patent Document 2: Sukhbaatar, S., Fergus, R., et al. Learning multiagent communication with backpropagation. In NIPS'16, 2016.
Non-Patent Document 3: Wu, Shih-Hung and Soo, Von-Wun. Game theoretic reasoning in multi-agent coordination by negotiation with a trusted third party. In Proceedings of the third annual conference on Autonomous Agents, pp. 56-61. ACM, 1999.
Non-Patent Document 4: Sandholm, Tuomas and Wang, XiaoFeng. (im) possibility of safe exchange mechanism design. In Eighteenth national conference on Artificial intelligence, pp. 338-344. American Association for Artificial Intelligence, 2002
Non-Patent Document 5: Agogino, Adrian K and Tumer, Kagan. Quicr-learning for multi-agent coordination. In Proceedings of the National Conference on Artificial Intelligence, volume 21, pp. 1438, 2006.

However, if rewards are distributed evenly, each agent tries to free ride on the achievements of others, resulting in a tragedy of the commons. I had a problem.

In order to solve the above problems, a first aspect of the present invention provides an information processing apparatus. The information processing device provides node input information to each node device of a node group including a plurality of node devices functioning as nodes forming an information network, and node output as a result of processing the node input information by each of the node devices. a network processing unit that acquires information, the network processing unit that provides node output information acquired from an upstream node device in an information network as node input information for a downstream node device connected to the upstream node device; Be prepared. The information processing device reduces the error between the node output information acquired from the most downstream node device corresponding to the most downstream node in the information network and the actual observation value corresponding to the node output information. A learning processing unit may be provided for learning the information network by updating the information transfer weights of the . The information processing device includes a reward distribution unit that makes the node device on the receiving side that receives the node output information as the node input information pay for the node output information to the node device on the output side that outputs the node output information. Be prepared.

The information network is a neural network and the nodes may correspond to neurons of the neural network.

The learning processing unit propagates the error between the actual observed value corresponding to the node output information and the node output information from the most downstream node device to the upstream node device, and based on the propagated error The weight of information transmission between node devices may be updated.

The information processing device may further include an investment execution unit that executes investment based on node output information acquired from the most downstream node device.

The investment execution unit executes investment based on the node output information obtained from the most downstream node device in response to obtaining an investment execution request from the investment terminal, and invests at least part of the profit obtained from the investment. distributed to terminals.

The reward distribution unit may distribute at least part of the profit obtained from the investment to the node device that has output the node output information used for the investment to the investment terminal.

The reward distribution unit may determine the consideration for the node output information based on the weight between the node device on the receiving side and the node device on the output side.

The reward distribution unit may cause the node device on the receiving side to pay for the node output information for the node device on the output side using cryptocurrency using blockchain.

The reward distribution unit may receive the public key B from the node device on the receiving side and transmit it to the node device on the output side in response to the payment of the consideration. The reward distribution unit may receive public key B (encryption key A') obtained by encrypting decryption key A' with public key B from the node device on the output side and transmit it to the node device on the receiving side. The reward distribution unit receives, from the output-side node device, encryption key A (node output information m) obtained by encrypting node output information m with encryption key A corresponding to decryption key A' from the output-side node device. , may be transmitted to the receiving node device.

The information processing device may further include a processing acquisition unit that acquires the type of node input information input by each node device and the type of node output information output by each node device from each of the plurality of node devices.

The information processing device connects the plurality of node devices based on the types of node input information and the types of node output information of the plurality of node devices acquired by the process acquisition unit, thereby forming an information network. It may further include a network generator for performing.

The network generating unit determines whether the type of node output information of the first node device and the type of node input information of the second node device match, and if both match, the first node device A second node device may be connected downstream of the .

An information processing system including the information processing device described above and a plurality of node devices may be provided. The information handling system may further comprise an investment terminal.

A second aspect of the present invention provides an information processing method. The information processing method provides node input information to each node device of a node group including a plurality of node devices functioning as nodes forming an information network, and node output as a result of processing the node input information by each of the node devices. It may have a network processing step of obtaining information and providing node output information obtained from an upstream node device as node input information of a downstream node device connected to the upstream node device. The information processing method includes node output information obtained from the most downstream node device corresponding to the most downstream node in the information network, and the node output information and the actual observed value corresponding to the node output information. There may be a training phase for training the information network by updating the signaling weights of . The information processing method includes a reward distribution step in which the node device on the receiving side that receives the node output information as node input information pays the consideration for the node output information to the node device on the output side that outputs the node output information. may have

A third aspect of the present invention provides an information processing program. The information processing program may be executed by a computer and cause the computer to function as a network processing unit. The information processing program may cause the computer to function as a learning processing unit. The information processing program may be executed by a computer and cause the computer to function as a reward distribution unit.

It should be noted that the above summary of the invention does not list all the necessary features of the invention. Subcombinations of these feature groups can also be inventions.

1 shows an overall overview of an information processing system 10 according to this embodiment. 1 shows an example of an apparatus configuration of an information processing system 10 according to this embodiment. 2 shows a processing flow of the information processing system 10 according to this embodiment. 4 shows a processing flow of each node device 32 in S300 of FIG. 3 according to the present embodiment. 5 shows an overview of the processing in S310 of FIG. 4 according to the present embodiment; FIG. 5 shows a subflow of S310 in FIG. 4 according to a modification of the present embodiment; FIG. 1 shows an example of a block chain used for cryptocurrency according to this embodiment. 1 shows an example of the configuration of a computer 2200 according to this embodiment.

Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. Also, not all combinations of features described in the embodiments are essential for the solution of the invention.

FIG. 1 shows an overview of an information processing system 10 according to this embodiment. In this embodiment, the information processing system 10 generates revenue by using an information source 20, a node group including a plurality of nodes 30 that process information obtained from the information source 20, and information output from the node group. It comprises an actuator 40, an investor 41 investing in the actuator 40, and a crowdsource platform 50 that operates them on the network.

The information processing system 10 provides a framework in which a plurality of nodes 30 and actuators 40 form an information network in which information processing is performed in cooperation, and each receives rewards for information processing independently. As a result, the information processing system 10 may implement MARL with each of the plurality of nodes 30 and actuators 40 as agents.

The information source 20 provides various information accessible on a network such as the Internet, various sensor data, etc. free of charge or charged. Information source 20 may be implemented by one or more server devices, databases, or other computers.

For example, the information source 20 is a server device that provides information on user attributes (age, gender, occupation, affiliation, etc.) and/or posted content in a social networking service (SNS) or a text posting service on the Internet. / A server device that provides a video sharing service, a server device that provides an Internet dictionary/dictionary service, a server device that provides web camera shooting information, a server device that provides financial information, a database of an academic institution that provides academic papers, etc. It may be a public agency database that provides meteorological data (for example, past weather information, temperature information, and/or atmospheric pressure information, etc.), a public agency database that provides statistical data (for example, demographic data), etc. .

Further, for example, the information source 20 includes environmental sensors that detect environmental information (for example, temperature sensors installed at multiple points, atmospheric pressure sensors, rainfall sensors, illuminance sensors, magnetic force sensors, heat sensors, image sensors, vibration sensors, voice sensor, etc.), position sensors that detect position information (e.g., GPS data of vehicles and people, speed data, acceleration data, etc.), traffic detection sensors (e.g., ETC gates, etc.), and/or other sensors, Alternatively, it may be a server device or the like that provides observation data of these sensors.

A plurality of nodes 30 are connected to each other either directly through a network or through a crowdsourced platform 50 . Each node 30 acquires node input information, performs an operation on the node input information, and outputs the operation result to another node 30 as node output information. In one example, the information network formed by nodes 30 is a neural network, and nodes 30 may correspond to neurons of the neural network.

Of the plurality of nodes 30, the nodes 30 (nodes n ₁₁ to n ₁₃ in the figure) located on the most upstream side (information source 20 side) are feature extractors for extracting features of information provided from the information source 20. It can function as a vessel. For example, the nodes n ₁₁ to n ₁₃ extract features of information provided by the information source 20 and convert them into data formats that can be used by the subsequent nodes n ₂₁ to n ₂₃ .

As an example, the node n ₁₁ performs text mining on text data of one or more websites to be monitored (for example, news articles in the financial field) using a conventionally known technique to extract text features. For example, the node n ₁₁ extracts features represented by scalar values, vector values, tensor values, etc. (hereinafter collectively referred to as “vector values, etc.”) from the mined text, and It may be output as node output information.

As another example, the node n ₁₁ accesses the meteorological database of the Japan Meteorological Agency and statistically processes the acquired meteorological data (for example, weather information, temperature information, and/or atmospheric pressure information at a specific point). A feature represented by a vector value or the like may be extracted from the meteorological data by conversion processing or the like, and the extracted feature may be output as node output information.

Further, the most upstream node 30 may output continuous time-series vector values or the like as node output information. For example, the most upstream node 30 samples constantly changing sensor data output from one or more sensors at predetermined time intervals (for example, every second, minute, hour, or day). Stream data such as vector values may be generated from the values and output as node output information.

The most upstream node 30 may be operated by the same or a separate independent business operator or individual. In addition, the node 30 on the most upstream side can be used by each business operator, etc., according to their area of expertise or interest, without worrying about the trends of other business operators, or the like, or the node 30 on the most upstream side. You can output information that is not much different from The most upstream node 30 may be realized by a neural network.

Nodes 30 (nodes n ₂₁ to n ₂₃ and nodes n ₃₁ to n ₃₃ located downstream of the most upstream node 30 (nodes n ₁₁ to node n ₁₃ ) among the plurality of nodes 30: “Downstream side Each node 30 acquires feature information from the upstream node 30, processes the information, and provides it to the downstream node 30.

For example, nodes n ₂₁ to n ₂₃ are connected downstream of nodes n ₁₁ to n ₁₃ , obtain node output information from nodes n ₁₁ to n ₁₃ as node input information, and perform calculations on The nodes n ₂₁ to n ₂₃ output the computation results as node output information to the nodes n ₃₁ to n ₃₃ further downstream. As an example, the node n ₂₁ calculates the prediction result of the future price of crude oil based on the node input information (for example, information related to weather data) obtained from the nodes n ₁₁ to n ₁₃ , and includes the prediction result Output node output information.

Here, the downstream node 30 generates node output information based on the result of weighting the calculation result generated based on the node input information from each node 30 connected in the previous stage with the weight between each node 30 can be output. For example, node n ₂₁ computes w ₁₁₋₂₁ ×f ₂₁ (n _1,1 )+w ₁₂₋₂₁ ×g ₂₁ (n _1,2 )+w ₁₃₋₂₁ ×h ₂₁ (n _1,3 )+b ₂₁ and may be output as node output information. Here, "x" may be an arithmetic symbol indicating the product of scalar values or the inner product of vector values or tensor values.

where n _i,j denotes node input information from node n _ij which may be a vector value, etc., and f _ij (n _i−1,x ), g _ij (n _i−1,y ), and , h _ij (n _i−1,z ) perform operations on node input information from node n _i−1,x , node n _i−1,y , and node n _i−1,z at node n _ij . where w _ij-ab represents the weight between node n _ij and node n _ab and b _ij represents a predetermined bias. w _ij-ab and b _ij may be vector values or the like. The node 30 only performs operations f _ij (n _i−1,x ), g _ij (n _i−1,y ), h _ij (n _i−1,z ), and the weighted node output information is the cloud It may be computed at source platform 50 . Alternatively, node 30 may perform the weighting process itself.

Of the downstream nodes 30 , the most downstream nodes (nodes n ₃₁ to n ₃₃ : sometimes referred to as “the most downstream node 30 ”) output node output information to the actuator 40 . For example, nodes n ₃₁ to n ₃₃ acquire node output information from nodes n ₂₁ to n ₂₃ as node input information, perform operations on the node input information, and weight the operation results. Output to the actuator 40 as node output information.

As an example, the node n ₃₁ calculates the future stock price prediction result based on the node input information obtained from the nodes n ₂₁ to n ₂₃ (for example, information related to the prediction result of the future price of crude oil), Node output information containing results may be output. The nodes n ₃₂ to n ₃₃ also acquire the same or different information as that of the node n ₃₁ , execute their own calculations, and output node output information based on the prediction results of future stock prices. good.

The downstream node 30 may be operated by an independent same or separate business operator or individual such as a data scientist, and each business operator may be aware of the trends of other business operators according to their areas of expertise or interests. You may independently process the information collected without Also, the downstream node 30 may output the same information as the other downstream nodes 30 . The downstream node 30 may output node output information including stream data. The downstream node 30 may be realized by a neural network.

Actuators 40 generate revenue using information received from one or more nodes 30 . For example, actuators 40 may generate revenue and receive rewards for investors 41 that provide investment capital. Actuator 40 may be operated by the same or a separate independent business or individual, such as a data scientist. Actuator 40 may be realized by a neural network.

The actuator 40 originally functions as part of a node in an information network, but is called an "actuator" to distinguish it from other nodes in terms of function. Although two actuators 40A and 40B are shown in the drawing, these are collectively referred to as actuator 40. As shown in FIG.

For example, the actuator 40A may generate an investment plan with the highest profit based on the node input information obtained from the node n ₃₁ (for example, prediction results of future stock prices of stocks of a plurality of specific stocks). As an example, the actuator 40A may generate an investment plan to purchase the stock with the largest stock price increase rate among the multiple stocks.

Also, for example, the actuator 40A may predict the amount of profit to be obtained in the future from the investment plan, and output the predicted amount of profit as node output information. As a result, the actuator 40A can make a profit if the prediction result of the profit amount of the investment plan is correct.

As an example, the actuator 40 may have one or more deep Q-Networks that approximate the action-value function Q, and output the investment plan by learning the deep Q-Network. Specifically, the actuator 40 inputs the state s _t at time t (for example, stock prices, technical indicators, total assets, and/or the number of shares held at time t) into one multilayer neural network, and at time t An action value Q(s _t , α _t ) that indicates the value of the action α _t may be output. Here, for example, the actuator 40A obtains the node input information obtained from the node n ₃₁ as at least a part of the state s _t , the action α _t as the investment plan, and the action value Q(s _t , α _t ) as the node You may output it as output information.

The actuator 40 receives the state s _t , performs α _t with the highest action value, receives the reward r _t and the next state s _t+1 from the market, and stores them as past experience during learning of the multi-layer neural network. Then, a part of them may be selected at random and mini-batch learning may be performed. The reward r _t may be a value according to the profit rate obtained by the action α _t . The action-value function Q may be learned by gradient descent. Such a method is described in, for example, Yuki Wada, Tomoharu Nagao, Construction of Stock Trading Strategy by Deep Learning, 6N-04, 79th National Convention of Information Processing Society of Japan, but is not limited thereto.

When connected to a plurality of nodes 30 like the actuator 40B, the actuator 40 has a plurality of (for example, two in the case of the actuator 40B) multilayer neural networks corresponding to each of the plurality of nodes, Node input information from each connected node 30 may be used as at least part of the state s _t to train each multilayer neural network.

Then, for example, actuator 40B may output an investment plan and node output information based on the outputs of multiple multilayer neural networks. As an example, actuator 40B may determine action α _t by majority vote of action values Q(s _t ,α _t ) weighted by w _32-B and w _33-B . Further, the actuator 40B may output the weighted average of the action value Q(s _t ,α _t ) corresponding to the determined action α _t by w _32-B and w _33-B as node output information.

An investor 41 invests in the actuator 40 . The investor 41 provides investment funds to the actuator 40 to generate profits from the investment plan in the actuator 40, and collects part of the profits. For example, investor 41A invests in actuator 40A and investor 41B invests in actuator 40B. Investor 41 may be operated by an independent business operator or individual such as a financial institution, company, organization, or individual investor.

Information exchanged between the plurality of nodes 30 and between the nodes 30 and the actuators 40 in the information processing system 10 may be traded. For example, actuator 40A may pay node n ₃₁ for future stock price prediction results. As a result, the actuator 40A may secure the amount obtained by subtracting the consideration paid to the node n ₃₁ from the received reward as profit.

The node n ₃₁ may pay for obtained node input information (eg, forecasted future price of crude oil) to generate node output information (eg, forecasted future stock price). For example, node n ₃₁ may pay node n ₂₁ . Node n ₃₁ may also pay consideration to node n ₂₂ and node n ₂₃ if it obtains node input information from these nodes. Similarly, the nodes n ₂₁ -n ₂₃ may pay for the node input information to the preceding nodes n ₁₁ -n ₁₃ . As a result, each node 30 may secure the amount obtained by subtracting the price paid to another node 30 from the price obtained by itself as a profit.

The crowdsource platform 50 is communicably connected to the nodes 30 and the actuators 40 and provides a platform for information transmission, consideration exchange, etc. between the nodes 30 and the actuators 40 . For example, the crowdsource platform 50 mediates information transmission between the information source 20 and the node 30, information transmission between the nodes 30, information transmission between the node 30 and the actuator 40, and other necessary information processing. .

Crowdsource platform 50 may also provide a user interface to nodes 30 and actuators 40 . For example, the crowdsource platform 50 may provide users of each node 30 and actuator 40 with search functionality for the nodes 30 within the information processing system 10 . For example, the crowdsource platform 50 stores information such as the processing content of each node 30 and its explanation, receives a request from a user of a certain node 30, searches for this information, and provides it to the user. good.

Accordingly, the crowdsource platform 50 can assist each node 30 to search and connect to another node 30, thereby allowing multiple nodes 30 to autonomously form and update an information network. Also, the crowdsource platform 50 may automatically form an information network from the processing content of each node 30 .

As described above, according to the information processing system 10 of the present embodiment, it is possible to realize an information network such as a neural network by combining a plurality of nodes managed by a plurality of businesses.

In FIG. 1, as the information network of the information processing system 10, some of the nodes 30 (nodes n ₁₁ to n ₁₃ ) are input nodes, and other parts of the node 30 (nodes n ₂₁ to n ₃₃ ) are two nodes. Although a four-layer neural network has been described with the intermediate nodes of the layers and the actuator 40 as the output node, it is not limited to this. For example, an arbitrary network of intermediate nodes may form between an input node and an output node. For example, in some embodiments, the nodes that make up the information network may not form distinctly separated layers, and some nodes may form a recursive network.

FIG. 2 shows an example of the device configuration of the information processing system 10 according to this embodiment. The information processing system 10 includes multiple node devices 32 , one or multiple investment terminals 42 , and an information processing device 100 . The nodes 30 and actuators 40 in FIG. 1 are realized by a plurality of node devices 32 , the investor 41 is realized by the investment terminal 42 , and the crowdsource platform 50 is realized by the information processing device 100 .

A plurality of node devices 32 perform information processing of each node in the information network. Here, some of the plurality of node devices 32 implement at least part of the functions of the node 30 in the information network. Another part of the plurality of node devices 32 implements at least part of the function of the actuator 40 .

The investment terminal 42 executes the investment in the information network. For example, investment terminal 42 executes an investment against an investment plan generated by actuator 40 .

The node device 32 and the investment terminal 42 (hereinafter both may be collectively referred to as “the node device 32, etc.”) are PCs (personal computers), tablet computers, smartphones, workstations, server computers, general-purpose computers, or the like. It may be a computer, or a computer system in which a plurality of computers are connected. For example, the node devices 32 and the like may be realized by multiple computers on a grid computing environment. As an example, the node device 32 or the like may form its own neural network with a plurality of computers. That is, each of the plurality of node devices 32 and the like functioning as nodes of the neural network may be realized by an individual neural network.

Also, the node device 32 and the like may be implemented by one or a plurality of virtual computer environments that can be executed within a computer. Alternatively, the node devices 32 and the like may be dedicated computers designed to implement the functions to be implemented by the node devices 32 and the like, or may be dedicated hardware implemented by dedicated circuits. The node device 32 and the investment terminal 42 may communicate with the information processing device 100 via a predetermined interface such as API (application program interface).

The information processing device 100 implements at least part of the functions of the crowdsource platform 50 . The information processing device 100 provides a platform that allows the node devices 32 and the like to cooperatively perform information processing while each of them independently obtains a reward for providing information and the like, thereby maximizing profits. The information processing device 100 includes a process acquisition unit 120 , a network generation unit 130 , a network processing unit 140 , an investment execution unit 150 , a learning processing unit 160 , a reward distribution unit 170 and a storage unit 180 .

The information processing apparatus 100 may be a computer such as a PC (personal computer), a tablet computer, a smart phone, a workstation, a server computer, or a general-purpose computer, or may be a computer system in which multiple computers are connected. For example, the information processing apparatus 100 may be realized by multiple computers on a grid computing environment. In this case, each of the multiple computers may implement one or more functions of the information processing apparatus 100 .

Further, the information processing apparatus 100 may be implemented by one or a plurality of virtual computer environments that can be executed within a computer. Alternatively, the information processing device 100 may be a dedicated computer designed to implement the functions to be implemented by the information processing device 100, or may be dedicated hardware realized by dedicated circuits.

The processing acquisition unit 120 may acquire the processing content of each node device 32 . For example, the process acquisition unit 120 may acquire the type of node input information input by each node device 32 and the type of node output information output by each node device 32 from each of the plurality of node devices 32 .

The network generation unit 130 may form an initial information network based on the processing content of each node device 32 acquired by the processing acquisition unit 120 . For example, the network generation unit 130 connects the plurality of node devices 32 based on the type of node input information and the type of node output information of the plurality of node devices 32 acquired by the process acquisition unit 120. may form an information network.

Also, the network generator 130 may assist the plurality of node devices 32 to autonomously form an information network. For example, the network generator 130 may provide the node device 32 with an interface for searching for other node devices 32 and connecting to the other node devices 32 .

The network processing unit 140 mediates information transmission between the plurality of nodes 30 and between the nodes 30 and the actuators 40 in the information network. For example, the network processing unit 140 provides node input information to each of the node devices 32 forming a node group of the information network, and acquires node output information as a result of processing the node input information from each of the node devices 32. good.

The network processing unit 140 may transmit information from the upstream node 30 to the downstream node in the information network. For example, the network processing unit 140 may provide node output information acquired from the upstream node device 32 in the information network as node input information for the downstream node device 32 connected to the upstream node device 32 .

The network processing unit 140 may mediate information transmission between the node devices 32 by providing an API (Application Program Interface) to the node devices 32 . For example, each node device 32 may exchange information with other node devices 32 by calling an API provided by the network processing unit 140 .

The investment execution unit 150 executes investment based on information obtained from the information network. Specifically, the investment execution unit 150 acts as an intermediary between the actuator 40 and the investor 41 and invests the funds invested by the investor 41 in the investment plan generated by the actuator 40 . As a result, the investment execution unit 150 can make a profit from the information processed by the information network.

For example, the investment execution unit 150 may execute investment based on node output information acquired from the most downstream node device 32 (for example, the node device 32 corresponding to the actuators 40A and 40B in FIG. 1) of the information network. As an example, the investment execution unit 150 executes investment based on an investment plan (for example, purchase of stocks of the specific brand at a specific time and a specific price) generated by the most downstream node device 32 (for example, the actuator 40A). .

The investment execution unit 150 may execute the investment and distribute at least part of the profit obtained from the investment to the investment terminal 42 in response to the investment execution request acquired from the investment terminal 42 . For example, when the investment terminal 42 purchases a fund corresponding to the actuator 40A, the investment execution unit 150 uses the purchase price as investment funds to execute the investment plan generated by the actuator 40A, and obtains At least a portion of the profits earned may be distributed to investment terminals 42 .

The learning processing unit 160 learns an information network realized by a plurality of node devices 32 . For example, the learning processing unit 160 acquires node output information (for example, the most downstream node device 32 between the node devices 32 so as to reduce the error between the actual observed value corresponding to the node output information (for example, the amount of future income from the investment plan generated by the investment plan) and the actual observed value (for example, the amount of income actually obtained from the investment plan) The information network may be trained by updating the information transfer weights.

As an example, the learning processing unit 160 propagates the error between the actual observed value corresponding to the node output information and the node output information from the most downstream node device 32 to the upstream node device 32, and propagates the error. The weight of information transmission between the node devices 32 may be updated based on the calculated error. Such a learning method is also called an error propagation method (or back propagation) in neural networks. As a result, the learning processing unit 160 can improve the accuracy of prediction based on the node output information while repeatedly updating the weights.

The remuneration distribution unit 170 distributes profits obtained from the investment of the investment execution unit 150 . For example, the reward distribution unit 170 distributes the profit obtained from the investment to the node device 32 (for example, the node device 32 corresponding to the actuators 40A and 40B in FIG. 1) that outputs the node output information used for the investment. At least a portion may be distributed.

Further, the reward distribution unit 170 causes the node device 32 on the receiving side that receives the node output information as the node input information to pay the consideration for the node output information to the node device 32 on the output side that outputs the node output information. you can As a result, the reward distribution unit 170 propagates the reward from the most downstream node device 32 corresponding to the actuator 40 to the upstream node device 32 corresponding to the nodes n ₃₃ to n ₁₁ through the information trading value. may be distributed to

Here, the reward distribution unit 170 determines the value of the node output information based on the weight between the node device 32 on the receiving side and the node device 32 on the output side (for example, the weight updated by the learning processing unit 160). You can decide. As a result, more rewards can be distributed to node devices 32 that are highly important in the information network (for example, that contribute greatly to profits from investment).

The storage unit 180 may include at least one of volatile and/or non-volatile storage devices that store information necessary for the operation of the information processing device 100 . For example, the storage unit 180 stores a program or the like that implements at least part of the processing acquisition unit 120, the network generation unit 130, the network processing unit 140, the investment execution unit 150, the learning processing unit 160, and the reward distribution unit 170. you can

Also, the storage unit 180 may at least temporarily store information generated in these operations. For example, the storage unit 180 may store node input information and/or node output information processed by the network processing unit 140 . As another example, the storage unit 180 may store information and the like acquired by the process acquisition unit 120 .

FIG. 3 shows the processing flow of the information processing system 10 according to this embodiment. In this embodiment, the information processing system 10 learns the information network and distributes the profit obtained from the information network by executing the processes from S100 to S800. An example in which the information network is a neural network and the nodes are neurons will be mainly described below, but the present embodiment is not limited to this.

First, in S100, the processing acquisition unit 120 of the information processing device 100 acquires the processing contents of the plurality of node devices 32 functioning as the plurality of neurons of the neural network. For example, the process acquisition unit 120 acquires the type of node input information and the type of node output information from each node device 32 . As an example, the process acquisition unit 120 acquires weather information, temperature information, and atmospheric pressure data of a specific point as types of node input information from the node device 32 serving as the node n ₂₁ in FIG. Obtain the future price of crude oil as the type.

Next, in S200, the network generation unit 130 of the information processing device 100 forms a neural network based on the type of node input information and the type of node output information of the node device 32 acquired in S100. For example, the network generation unit 130 determines whether the type of the node output information of the first node device and the type of the node input information of the second node device match. By connecting a node corresponding to a second node device downstream of a node (ie, neuron) corresponding to one node device, the neurons may be connected to each other, thereby automatically forming a neural network.

As an example, the network generation unit 130 generates the type of node output information of the _{node n11} in FIG. information, temperature information, and atmospheric pressure information), the node _n21 may be connected to the downstream side of the node _n11 . Also, as an example, the network generation unit 130 may generate the type of node output information (for example, future crude oil price) of node n ₂₁ in FIG. 1 and the type of node input information (for example, future crude oil price) of node n ₃₁ . may be connected to the _downstream side of node n ₂₁ in response to determining that . Further, the network generation unit 130 determines that the type of node output information of the node n ₃₁ in FIG. Actuator 40A may be connected to the downstream side of node n ₃₁ in response to the determination of the match.

In this way, the network generator 130 forms an information network such as a neural network by connecting nodes (or actuators) that want certain information to nodes that output such information.

In S200, the user of each node device 32 may manually form at least part of the neural network. For example, the network generator 130 may provide a search function for the nodes 30 in the information processing system 10 to users of each node 30 and actuator 40 (ie, node device 32). Thereby, the user of each node device 32 may find another node 30 that provides the information desired by his node device 32 and connect himself to it.

Next, at S300, the information processing system 10 performs prediction using the neural network formed at S200. For example, the node devices 32 corresponding to the nodes n ₁₁ to n ₁₃ shown in FIG. get. The node device 32 corresponding to the nodes n ₁₁ to n ₁₃ may acquire information by directly accessing the information source 20, or via the network processing unit 140 of the information processing device 100 functioning as the crowd source platform 50, Information may be obtained from information source 20 .

The node device 32 corresponding to the nodes n ₁₁ to n ₁₃ performs arithmetic processing on the acquired node input information and outputs the processing result to the network processing section 140 .

The network processing unit 140 supplies the node output information of the nodes n ₁₁ to n ₁₃ thus obtained as node input information to the nodes n ₂₁ to n ₂₃ connected downstream of these nodes. The network processor 140 may supply node input information only to downstream nodes connected to upstream nodes. For example, in FIG. ₁ , nodes n ₁₁ to n ₁₃ are all connected to nodes n ₂₁ to n ₂₃ , respectively _. , the node output information of node _n11 may be provided only to node _n21 .

Next, the node device 32 corresponding to the nodes n ₂₁ to n ₂₃ acquires node output information from the nodes n ₁₁ to n ₁₃ from the network processing section 140 as node input information. The node devices 32 corresponding to the nodes n ₂₁ to n ₂₃ perform arithmetic processing on the received node input information and supply it to the network processing section 140 .

The network processing unit 140 may weight the obtained processing results by the method described above. For example, the network processing unit 140 acquires f ₂₁ (n _1,1 ), g ₂₁ (n _1,2 ), and h ₂₁ (n _1,3 ) from the node device 32 corresponding to the node n ₂₁ . , w ₁₁₋₂₁ ×f ₂₁ (n _1,1 )+w ₁₂₋₂₁ ×g ₂₁ (n _1,2 )+w ₁₃₋₂₁ ×h ₂₁ (n _1,3 )+b ₂₁ , which is the node n ₂₁ node output information.

Note that the weighting may be performed by the node device 32 itself. For example, the node device 32 may receive weight information related to itself from the network processing unit 140, perform weighting processing, and provide the result to the network processing unit 140 as node output information.

The network processing unit 140 may use the weighted result as the node output information of each node. Node output information may include indices, etc. that are predicted based on information from information sources 20 . For example, node output information for node n ₂₁ may include information related to the expected outcome of future prices of crude oil. The network processing unit 140 may similarly compute the node output information of the nodes n ₂₂ to n ₂₃ and supply this to the downstream nodes n ₃₁ to n ₃₃ as node input information.

Next, the node device 32 corresponding to the nodes n ₃₁ to n ₃₃ acquires node output information from the nodes n ₂₁ to n ₂₃ from the network processing unit 140 as node input information. The node device 32 corresponding to the nodes n ₃₁ to n ₃₃ performs arithmetic processing on the received node input information and supplies the result to the network processing section 140 . The network processing unit 140 performs weighting processing on the obtained calculation results and uses the result as node output information of each node.

For example, the network processing unit 140 outputs, as the node output information of the node n ₃₁ , the prediction result of the future stock price of the brand A, the prediction result of the future stock price of the brand B, and the information related to the prediction result of the future stock price of the brand C. You can The network processing unit 140 may supply the actuator 40A with information related to the prediction results of future stock prices of the specific stocks A to C as node input information. The network processing unit 140 may similarly supply the node output information of the nodes n ₃₂ to n ₃₃ to the actuator 40B as node input information.

The node device 32 corresponding to the actuator 40A and the actuator 40B performs arithmetic processing on the received node input information and supplies the result to the network processing section 140. FIG. The node devices 32 corresponding to the actuators 40A and 40B may output, for example, the above-described action _αt and action value Q as calculation results. For example, the network processing unit 140 weights the obtained calculation results and uses at least part of the result as the node output information of each actuator 40 .

As a result of the weighting process, the network processing unit 140 acquires the investment plan with the highest profit as node output information, and supplies it to the investment execution unit 150 . The actuator 40 may receive its own weight from the network processing unit 140 and perform weighting by itself.

The investment plan may include execution conditions such as, for example, the brand of stock to be purchased, purchase timing, and price designation. The network processing unit 140 may supply the learning processing unit 160 with the amount of profit to be obtained in the future from the investment plan that will maximize the profit as the node output information of the actuator 40A.

Thus, at S300, the information processing system 10 generates an investment plan that is expected to generate a high profit and a predicted profit from the investment plan based on public information provided from the information source 20 and the like.

Next, at S400, the investment execution unit 150 accepts an investment from the investment terminal 42. FIG.

The investment execution unit 150 may sell the fund corresponding to the investment plan generated in S300 to the investment terminal 42 . Investment terminal 42 may purchase funds that are sold. For example, the investment terminal 42 remits the investment fund to the investment execution unit 150 as the payment for the fund, and also transmits an investment execution request designating the fund. The investment execution unit 150 acquires an investment execution request from the investment terminal 42 .

In S600, the investment execution unit 150 executes investment based on the investment execution request. For example, the investment execution unit 150 may invest at least a portion of the fund price as investment funds in an investment plan corresponding to the fund specified in the investment execution request. As an example, the investment execution unit 150 may access a server device of a securities company via a network and execute an investment corresponding to an investment plan (for example, purchase of stocks of a specific brand) with investment funds.

After that, the investment execution unit 150 collects the profit obtained from the investment. For example, the investment execution unit 150 detects that the price of the purchased stock has exceeded a predetermined price (for example, the scheduled price set in the investment plan) and/or that a predetermined time has passed. If necessary, sell the purchased shares to recover the invested funds. As an example, the investment execution unit 150 accesses the server device of the securities company via the network and sells the purchased issue.

Here, the investment executing section 150 may distribute at least part of the profit obtained from the investment to the investment terminal 42 . For example, the investment execution unit 150 may distribute the profit from the investment (the sale price of the stock of the specific stock minus the purchase price) to the investment terminal 42 that purchased the fund. The investment execution unit 150 may distribute the profits to the account assigned to the user of the investment terminal 42 . Also, the investment execution unit 150 may distribute profits to the account of the user of the investment terminal 42 in cryptocurrency (eg, bitcoin or other virtual currency) using blockchain.

The account may be an external account such as a bank account. In this case, the investment execution unit 150 may access a network of a financial institution such as a bank and deposit the funds into the distribution target account. Also, the account may be a virtual account within the information processing system 10 . In this case, the investment execution unit 150 may increase the account balance by the amount of the distributed profit and remit cryptocurrency to the investment terminal 42 in response to a request from the user of the investment terminal 42 .

The investment execution unit 150 may distribute to the user of the investment terminal 42 an amount obtained by deducting a certain percentage as a fee from the income and/or the investment funds from the income from the investment. For example, the investment execution unit 150 deducts a certain percentage of the investment fund (2% as an example) and a certain percentage of the investment income (20% as an example) as a fee from the investment income and distributes it to the user of the investment terminal 42. you can

The remuneration distribution unit 170 distributes part of the profit obtained from the investment to the information network. For example, the reward distribution unit 170 distributes the profit obtained from the investment to the node device 32 (for example, the node device 32 corresponding to the actuators 40A and 40B in FIG. 1) that outputs the node output information used for the investment. Distribute at least a portion. For example, the reward distribution unit 170 distributes at least part of the commission deducted in the distribution to the investment terminal 42 to the actuators 40A and 40B as consideration for the node output information.

Here, the node devices 32 corresponding to the actuators 40A and 40B also pay for the node output information to the node devices 32 on the upstream side (for example, the node devices 32 corresponding to the nodes n ₃₁ to n ₃₃ ). Therefore, the actuator 40 can secure the amount obtained by subtracting the price of the information from the distributed reward as profit.

In S700, the learning processing unit 160 calculates the error between the prediction by the neural network and the actual observed value. For example, the learning processing unit 160 calculates the difference between the predicted value predicted in S300 (for example, the amount of profit predicted by the investment plan) and the actual observed value (for example, the amount of profit actually obtained by the investment plan). , is calculated as an error.

In S800, the learning processing unit 160 propagates the error calculated in S700 from the most downstream node device 32 (for example, the actuators 40A and 40B) to the upstream node device 32, and based on the propagated error to update the weight of information transmission between the node devices 32 . The learning processing unit 160 may update the weights by a method conventionally known in the error propagation method of neural networks. As a result, the learning processing unit 160, for example, obtains the weights w ₁₁ — 21 to w 11 — _{23 ,} w ₁₂ — 21 to w 12 — 23, w ₁₃ — 21 to w 13 — _{23, w 21 — 31 to w 21 — 33 ,} w ₂₂ — 31 to w ₂₂ — ₃₃ , w _{23_31} to w _{23_33} , w _{31_A} , w _{32_B} and w _{33_B} may be updated.

As a result, the learning processing unit 160 can improve the prediction accuracy of the neural network by updating the weights so as to reduce the error. As will be described later, the consideration for node output information may be determined based on the weights among the node devices 32 . As a result, in the neural network, larger rewards can be distributed to the node devices 32 that output important information that affects prediction (that is, nodes that are weighted more heavily).

After that, the information processing system 10 may return to S300 and repeat the loop of S300-S800. Accordingly, the information processing system 10 may repeatedly use the neural network and update the neural network.

FIG. 4 shows the processing flow of each node device 32 in S300 of FIG. 3 according to this embodiment. In S300 of FIG. 3, each node device 32 may execute the processes of S310 to S350.

First, in S310, the node device 32 acquires node input information. For example, the node device 32 acquires node output information from another node device 32 connected upstream of the node device 32 as node input information from the network processing unit 140 of the information processing device 100 . The node device 32 may acquire node output information in the form of vector values or the like. As an example, the node device 32 corresponding to the node n ₂₁ may acquire, as node input information, vector values including past weather information, temperature information, and/or atmospheric pressure information.

The node device 32 on the receiving side that receives the node input information may pay consideration to the node device 32 on the output side that outputs the corresponding node output information when acquiring the node input information. That is, the node device 32 may purchase node input information. Details of the purchase process are described in FIGS.

In addition, the node device 32 on the most upstream side in the neural network (for example, the node device 32 corresponding to the node n ₁₁ ) accesses the information source 20 (for example, a weather database) or the like, and obtains past weather information, temperature information, and / Alternatively, information such as atmospheric pressure information may be obtained as node input information. Alternatively, the most upstream node device 32 may acquire information contained in the information source 20 via the network processing unit 140 of the information processing device 100 .

Next, in S330, the node device 32 performs arithmetic processing on the node input information acquired in S310. For example, the node device 32 performs a linear operation on node input information such as vector values. As an example, the node device 32 corresponding to the node n ₂₁ may compute f ₂₁ (n _1,1 ), g ₂₁ (n _1,2 ), and h ₂₁ (n _1,3 ). f _ij (n _i-1,x ), g _ij (n _i-1,y ), and h _ij (n _i-1,z ) are the nodes n _{i -1,x} , node It represents the operation for each of the node input information from n _i-1,y and node n _i-1,z . f _ij (n _i-1,x ), g _ij (n _i-1,y ) and h _ij (n _i-1,z ) can be arbitrary functions, for example linear functions good.

Further, in the neural network, the node device 32 on the most upstream side (for example, the node device 32 corresponding to the node _n11 ) converts the information acquired from the information source 20 in S310 into a vector value or the like instead of performing the above calculation in S330. can be converted to Thereby, the most upstream node device 32 functions as a feature extractor.

Next, at S<b>350 , the node device 32 outputs the calculation result at S<b>330 to the network processing unit 140 . The network processing unit 140 may generate node output information by weighting the output results from the node device 32 . For example, the network processing unit 140 receives f ₂₁ (n _1,1 ), g ₂₁ (n _1,2 ), and h ₂₁ (n _1,3 ) from the node device 32 corresponding to the node n ₂₁ and w ₁₁₋₂₁ ×f ₂₁ (n _1,1 )+w ₁₂₋₂₁ ×g ₂₁ (n _1,2 )+w ₁₃₋₂₁ ×h ₂₁ (n _1,3 )+b ₂₁ may be calculated. Note that the node device 32 may receive a weight associated with itself from the network processing unit 140, perform weighting processing instead of the network processing unit 140, and generate node output information.

The network processing unit 140 supplies the node output information to another node device 32 on the downstream side of the node device 32 that generated the node output information. For example, the network processing unit 140 sends the node output information of the node device 32 corresponding to the node n ₂₁ to each of the node devices 32 corresponding to the nodes n ₃₁ to n ₃₃ connected downstream of the node n ₂₁ , Supplied as node input information. Then, the processes of S310 to S350 are sequentially repeated in the node devices 32 corresponding to the nodes n ₃₁ to n ₃₃ and the node devices 32 corresponding to the actuators 40A and 40B.

FIG. 5 shows an overview of the processing in S310 of FIG. 4 according to this embodiment. In S<b>310 , the node device 32</b>A on the receiving side of the node input information transmits a consideration payment request to the reward distribution unit 170 of the information processing device 100 . The remuneration distribution unit 170 receives a consideration payment request from the node device 32A, and executes consideration payment processing from the node device 32A on the receiving side of the node input information to the node device 32B on the output side of the node output information. For example, the reward distribution unit 170 may remit an amount corresponding to the consideration from the account of the user of the node device 32A to the account of the user of the node device 32B.

The reward distribution unit 170 may decrease the account balance of the user of the node device 32A by the consideration and increase the account balance of the user of the node device 32B by the consideration. The reward distribution unit 170 may transfer cryptocurrency from the node devices 32A to 32B in response to a request from the user of the node device 32A. Also, the reward distribution unit 170 may access a network of a financial institution such as a bank and perform remittance processing from the account of the user of the node device 32A to the account of the user of the node device 32B.

In response to payment, the network processing unit 140 supplies node output information generated in advance from the node device 32B on the output side to the node device 32A on the receiving side. The node device 32A on the receiving side acquires the node output information from the node device 32B on the output side from the network processing unit 140 as node input information.

Here, the reward distribution unit 170 may determine the consideration for the node output information from the node device 32 based on the weight w _AB between the node device 32A on the receiving side and the node device 32B on the output side. . For example, if the weight w _AB is large, the contribution of information from the node device 32A to the node output information of the node device 32B is large. Therefore, the reward distribution unit 170 may increase the amount of consideration from the node device 32B to the node device 32A as the weight w _AB increases. For example, the reward distribution unit 170 may set the amount of consideration proportional to the weight w _AB .

Also, if the node device 32A pays the node device 32B an amount of money that is larger than the price it receives from another node device 32 downstream, a loss will occur. Therefore, the reward distribution unit 170 may determine that the total amount of consideration paid by the node device 32A to the upstream node devices 32 (including the node device 32B) does not exceed the consideration received by itself.

Also, the reward distribution unit 170 may determine the price through an auction of a plurality of node devices 32 . For example, when a plurality of downstream (receiving side) node devices 32A require node output information from the same upstream (output side) node device 32B, the reward distribution unit 170 distributes a plurality of downstream nodes A bid amount for each consideration may be received from device 32A and the highest bid amount may be determined as the consideration. In that case, the network processing unit 140 may supply target node output information only to the node device 32A that has made the highest bid.

Instead, the reward distribution unit 170 receives bid amounts for respective considerations from a plurality of receiving-side node devices 32A, the transaction is concluded (that is, the consideration becomes equal to or less than the bid amount), and the output side A price that can maximize the profit of the node device 32B may be determined as consideration. As an example, when the node device 32A ₁ bids 100, the node device 32A ₂ bids 80, and the node device 32A ₃ bids 10 for the node output information of the node device 32B, the reward The distribution unit 170 may set the amount of money 80 as the consideration. In this case, the network processing unit 140 may supply the node output information of the node device 32B to the node devices 32A- ₁ and 32A- ₂ .

If the node device 32B is implemented by a plurality of computers, the fee may be further distributed among the plurality of computers. For example, a plurality of computers that implement the node device 32B may distribute compensation according to their own calculation processing amounts. Such reward distribution may occur entirely within the computers, or the reward distribution component 170 may calculate the share of the consideration for the computers and distribute the share to each computer.

Here, the information processing system 10 may use a smart contract based on cryptocurrency to safely exchange compensation and information between the node device 32A on the receiving side and the node device 32B on the output side. Such an embodiment will be described below with reference to FIG.

FIG. 6 shows a subflow of S310 in FIG. 4 according to this embodiment. In the information processing system 10, the node device 32A may safely acquire the node input information from the node device 32B by performing the processes of S312 to S322 of FIG.

First, in S312, the reward distribution unit 170 of the information processing device 100 causes the node device 32A on the receiving side to pay for the node output information for the node device 32B on the output side in cryptocurrency using blockchain. A cryptocurrency that uses a blockchain may be a virtual currency such as bitcoin. A block in such a cryptocurrency blockchain may contain a transaction.

The reward distribution unit 170 may record payment of consideration from the node device 32A to the node device 32B in the transaction. The reward distribution unit 170 may further record the fulfillment details of the node device 32B (for example, information about the node output information output by the node device 32B, etc.) in the transaction.

Next, in S314, the node device 32A on the receiving side transmits its own public key A to the reward distribution unit 170 in response to payment of the consideration in S312. The node device 32A holds a private key A' corresponding to the public key A in a secret state. The reward distribution unit 170 receives the public key A from the node device 32A on the receiving side and transmits it to the node device 32B on the output side.

Next, in S316, the node device 32B on the output side encrypts the decryption key B' corresponding to the encryption key B held by itself with the public key A to generate a public key A (composite key B'), and obtains a reward. Send to distribution unit 170 . The reward distribution unit 170 receives the public key A (composite key B') from the node device 32B on the output side and transmits it to the node device 32A on the receiving side. In S312 to S316, the node device 32A on the receiving side receives the composite key B' of the node device 32B on the output side in exchange for the payment of the consideration by the blockchain. Therefore, it can be said that the node device 32A purchased the composite key B' from the node device 32B on the blockchain.

Next, in S318, the node device 32A on the receiving side decrypts the public key A (composite key B') with the secret key A' held by itself, and acquires the composite key B'.

Next, in S320, the node device 32B on the output _side receives the weight related to itself from the network processing unit 140 and generates its own node output information MB. The node device 32B on the output side generates an encryption key B (node output information M _B ) by encrypting the node output information M _B with the encryption key B, and transmits this to the reward distribution unit 170 . The reward distribution unit 170 receives the encryption key B (node output information M _B ) from the node device 32B and transmits it to the node device 32A.

Next, in S322, the node device 32A on the receiving side decrypts the encryption key B (node output information M _B ) with the decryption key B', and acquires the node output information M _B as node input information.

According to the embodiment described in FIG. 6, the node device 32A and the node device 32B can safely exchange node output information and consideration.

In the above embodiment, an example was described in which the node device 32A on the receiving side and the node device 32B on the output side exchanged consideration and information via the network processing unit 140. However, in the modified example of this embodiment, The node device 32A and the node device 32B may communicate directly without going through the network processing unit 140. FIG.

FIG. 7 shows an example of a block chain used for cryptocurrency according to this embodiment. A blockchain contains multiple blocks. FIG. 7 shows n blocks, block (1), block (2), block (3) . . . block (n-1), block (n)).

Each block may include a transaction, a hash of previously generated blocks, a hash of the transaction, and so on. For example, block (n) may include transaction (n), hash [block (n-1)], and hash [transaction (n)]. Here hash [*] represents a hash function. The transaction (n) may include the fulfillment details of the node device 32B (for example, information related to node output information output by the node device 32B).

According to the information processing system 10, a large number of agents (for example, nodes) work together in MARL, and a reward is distributed from the downstream side of the information network to the upstream side as consideration for information, so that a third party is not involved. Reward distribution can be realized. In particular, according to the information processing system 10 of the present embodiment, rewards can be distributed more heavily to agents who have contributed to achievements (for example, accurate prediction of earnings), so fairness in reward distribution is maintained. .

In the above description of the embodiment, an example in which the information processing system 10 is mainly used in the financial field was given, but the present invention is not limited to this. The information processing system 10 can be applied to all fields using distributed computing. For example, the information processing system 10 can be applied to various fields such as smart grids, manufacturing, insurance, and retail.

For example, information processing system 10 may be used to improve manufacturing efficiency in the manufacturing industry. In this case, the actuator 40 may be operated by, for example, a manufacturing business operator, and based on the information received from the node 30 on the upstream side, the amount of products to be manufactured, the composition ratio of the products to be manufactured, the country of manufacture, and other information. Altering the parameters may shorten production time, improve quality, and/or increase revenue (including cost reduction).

Actuator 40 may also predict, for example, weather, agricultural yields, and various other environmentally influenced events, and may generate revenue by selling the forecast data on behalf of a fund or the like. . Also, the actuator 40 may be a television operator, a newspaper company, a publisher, or other mass communication company, and based on the information received from the node 30 on the upstream side, generates or selects information to be provided by media. You can make a profit by doing so.

Various embodiments of the invention may be described with reference to flowchart illustrations and block diagrams, where blocks refer to (1) steps in a process in which operations are performed or (2) devices responsible for performing the operations. may represent a section of Certain steps and sections may be implemented by dedicated circuitry, programmable circuitry provided with computer readable instructions stored on a computer readable medium, and/or processor provided with computer readable instructions stored on a computer readable medium. you can Dedicated circuitry may include digital and/or analog hardware circuitry, and may include integrated circuits (ICs) and/or discrete circuitry. Programmable circuits include logic AND, logic OR, logic XOR, logic NAND, logic NOR, and other logic operations, memory elements such as flip-flops, registers, field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), etc. and the like.

Computer-readable media may include any tangible device capable of storing instructions to be executed by a suitable device, such that computer-readable media having instructions stored thereon may be designated in flowcharts or block diagrams. It will comprise an article of manufacture containing instructions that can be executed to create means for performing the operations described above. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, and the like. More specific examples of computer readable media include floppy disks, diskettes, hard disks, random access memory (RAM), read only memory (ROM), erasable programmable read only memory (EPROM or flash memory), Electrically Erasable Programmable Read Only Memory (EEPROM), Static Random Access Memory (SRAM), Compact Disc Read Only Memory (CD-ROM), Digital Versatile Disc (DVD), Blu-ray (RTM) Disc, Memory Stick, Integration Circuit cards and the like may be included.

The computer readable instructions may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state setting data, or object oriented programming such as Smalltalk, JAVA, C++, etc. language, and any combination of one or more programming languages, including conventional procedural programming languages, such as the "C" programming language or similar programming languages. good.

Computer readable instructions may be transferred to a processor or programmable circuitry of a general purpose computer, special purpose computer, or other programmable data processing apparatus, either locally or over a wide area network (WAN), such as a local area network (LAN), the Internet, or the like. ) and may be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, and the like.

FIG. 8 illustrates an example computer 2200 in which aspects of the invention may be implemented in whole or in part. The programs installed on the computer 2200 cause the computer 2200 to operate or function as one or more sections of the information processing device 100, node device 32, investment terminal 42, etc., associated with the embodiments of the present invention. or cause the operation or the one or more sections to be performed and/or cause the computer 2200 to perform the process or steps of the process according to embodiments of the present invention. Such programs may be executed by CPU 2212 to cause computer 2200 to perform certain operations associated with some or all of the flowchart and block diagram blocks described herein.

Computer 2200 according to this embodiment includes CPU 2212 , RAM 2214 , graphics controller 2216 , and display device 2218 , which are interconnected by host controller 2210 . Computer 2200 also includes input/output units such as communication interface 2222, hard disk drive 2224, DVD-ROM drive 2226, and IC card drive, which are connected to host controller 2210 via input/output controller 2220. there is The computer also includes legacy input/output units such as ROM 2230 and keyboard 2242 , which are connected to input/output controller 2220 through input/output chip 2240 .

CPU 2212 operates according to programs stored in ROM 2230 and RAM 2214, thereby controlling each unit. Graphics controller 2216 retrieves image data generated by CPU 2212 into itself, such as a frame buffer provided in RAM 2214 , and causes the image data to be displayed on display device 2218 .

Communication interface 2222 communicates with other electronic devices over a network. Hard disk drive 2224 stores programs and data used by CPU 2212 within computer 2200 . DVD-ROM drive 2226 reads programs or data from DVD-ROM 2201 and provides programs or data to hard disk drive 2224 via RAM 2214 . The IC card drive reads programs and data from IC cards and/or writes programs and data to IC cards.

ROM 2230 stores therein programs that are dependent on the hardware of computer 2200, such as a boot program that is executed by computer 2200 upon activation. Input/output chip 2240 may also connect various input/output units to input/output controller 2220 via parallel ports, serial ports, keyboard ports, mouse ports, and the like.

A program is provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from a computer-readable medium, installed in hard disk drive 2224 , RAM 2214 , or ROM 2230 , which are also examples of computer-readable medium, and executed by CPU 2212 . The information processing described within these programs is read by computer 2200 to provide coordination between the programs and the various types of hardware resources described above. An apparatus or method may be configured by implementing the manipulation or processing of information in accordance with the use of computer 2200 .

For example, when communication is performed between the computer 2200 and an external device, the CPU 2212 executes a communication program loaded into the RAM 2214 and sends communication processing to the communication interface 2222 based on the processing described in the communication program. you can command. The communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in a recording medium such as the RAM 2214, the hard disk drive 2224, the DVD-ROM 2201, or an IC card under the control of the CPU 2212, and transmits the read transmission data. Data is transmitted to the network, or received data received from the network is written to a receive buffer processing area or the like provided on the recording medium.

In addition, the CPU 2212 causes the RAM 2214 to read all or necessary portions of files or databases stored in external recording media such as a hard disk drive 2224, a DVD-ROM drive 2226 (DVD-ROM 2201), an IC card, etc. Various types of processing may be performed on the data in RAM 2214 . CPU 2212 then writes back the processed data to the external recording medium.

Various types of information, such as various types of programs, data, tables, and databases, may be stored on recording media and subjected to information processing. CPU 2212 performs various types of operations on data read from RAM 2214, information processing, conditional decision making, conditional branching, unconditional branching, and retrieval of information, as specified throughout this disclosure and by instruction sequences of programs. Various types of processing may be performed, including /replace, etc., and the results written back to RAM 2214 . In addition, the CPU 2212 may search for information in a file in a recording medium, a database, or the like. For example, if a plurality of entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored in the recording medium, the CPU 2212 determines that the attribute value of the first attribute is specified. search the plurality of entries for an entry that matches the condition, read the attribute value of the second attribute stored in the entry, and thereby associate it with the first attribute that satisfies the predetermined condition. an attribute value of the second attribute obtained.

The programs or software modules described above may be stored in a computer readable medium on or near computer 2200 . Also, a recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as a computer-readable medium, thereby providing the program to the computer 2200 via the network. do.

Although the present invention has been described above using the embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It is obvious to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the description of the scope of claims that forms with such modifications or improvements can also be included in the technical scope of the present invention.

The execution order of each process such as actions, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, the specification, and the drawings is particularly "before", "before etc., and it should be noted that they can be implemented in any order unless the output of the previous process is used in the subsequent process. Regarding the operation flow in the claims, the specification, and the drawings, even if the description is made using "first," "next," etc. for the sake of convenience, it means that it is essential to carry out in this order. not a thing

10 information processing system 20 information source 30 node 32 node device 40 actuator 41 investor 42 investment terminal 50 crowd source platform 100 information processing device 120 processing acquisition unit 130 network generation unit 140 network processing unit 150 investment execution unit 160 learning processing unit 170 reward Distribution unit 180 Storage unit

Claims (16)

Providing node input information to each node device of a node group including a plurality of node devices functioning as nodes forming an information network, and obtaining node output information as a result of processing the node input information by each of the node devices a network processing unit that provides node output information acquired from an upstream node device in the information network as node input information for a downstream node device connected to the upstream node device;
Information transmission between the node devices so as to reduce an error between node output information obtained from the most downstream node device corresponding to the most downstream node in the information network and an actual observed value corresponding to the node output information. a learning processing unit for learning the information network by updating the weights of
a reward distribution unit for causing a node device on the receiving side that receives the node output information as the node input information to pay a consideration for the node output information to a node device on the output side that outputs the node output information;
Information processing device. wherein said information network is a neural network and said nodes correspond to neurons of said neural network;
The information processing device according to claim 1 . The learning processing unit
Propagating an error between the actual observed value corresponding to the node output information and the node output information from the most downstream node device to the upstream node device, and based on the propagated error between the node devices update the signaling weights of
The information processing apparatus according to claim 1 or 2. executing an investment based on the node output information obtained from the most downstream node device;
further comprising an investment execution department;
The information processing apparatus according to any one of claims 1 to 3. The investment execution department
executing an investment based on the node output information obtained from the most downstream node device in response to obtaining an investment execution request from the investment terminal;
distributing at least a portion of the profits obtained from the investment to the investment terminal;
The information processing apparatus according to claim 4. The reward distribution unit
Distributing at least part of the profit obtained from the investment to a node device that outputs node output information used for the investment to the investment terminal;
The information processing device according to claim 5 . The reward distribution unit determines a consideration for the node output information based on the weight between the node device on the receiving side and the node device on the output side.
The information processing apparatus according to any one of claims 1 to 6. The reward distribution unit causes the receiving node device to pay the consideration for the node output information for the output node device with cryptocurrency using blockchain.
The information processing apparatus according to any one of claims 1 to 7. The reward distribution unit, in response to payment of the consideration,
receiving public key A from the node device on the receiving side and transmitting it to the node device on the output side;
receiving public key A (composite key B') obtained by encrypting decryption key B' with public key A from the node device on the output side and transmitting it to the node device on the receiving side;
An encryption key B (node output information M _B ) obtained by encrypting the node output information M _B with an encryption key B corresponding to the decryption key B′ from the output side node device is received from the output side node device. and send it to the node device on the receiving side,
The information processing apparatus according to claim 8 . further comprising a processing acquisition unit for acquiring, from each of the plurality of node devices, the type of the node input information input by each node device and the type of the node output information output by each node device;
The information processing apparatus according to any one of claims 1 to 9. connecting the plurality of node devices based on the type of the node input information and the type of the node output information of the plurality of node devices acquired by the processing acquisition unit, thereby establishing the information network; further comprising a network generator that forms
The information processing apparatus according to claim 10. The network generator,
Determining whether the type of the node output information of the first node device and the type of the node input information of the second node device match, and responding to the first node device in response to a match between the two connecting a node corresponding to the second node device downstream of the node that
The information processing device according to claim 11 . an information processing apparatus according to any one of claims 1 to 12;
the plurality of node devices;
An information processing system comprising 14. The information handling system of claim 13, further comprising an investment terminal. Giving node input information to each node device of a node group including a plurality of node devices functioning as nodes forming an information network, and acquiring node output information as a result of processing the node input information by each of the node devices a computer-implemented network processing step of providing node output information obtained from an upstream node device as node input information for a downstream node device connected to said upstream node device;
Information transmission between the node devices so as to reduce an error between node output information obtained from the most downstream node device corresponding to the most downstream node in the information network and an actual observed value corresponding to the node output information. a computer-implemented learning step of training the information network by updating the weights of
Executed by a computer, causing the node device on the receiving side that received the node output information as the node input information to pay for the node output information to the node device on the output side that output the node output information a reward distribution stage;
An information processing method comprising: executed by a computer, causing the computer to:
Providing node input information to each node device of a node group including a plurality of node devices functioning as nodes forming an information network, and obtaining node output information as a result of processing the node input information by each of the node devices a network processing unit that provides node output information acquired from an upstream node device in the information network as node input information for a downstream node device connected to the upstream node device;
Information transmission between the node devices so as to reduce an error between node output information obtained from the most downstream node device corresponding to the most downstream node in the information network and an actual observed value corresponding to the node output information. a learning processing unit for learning the information network by updating the weights of
a remuneration distribution unit that makes the node device on the receiving side that receives the node output information as the node input information pay the consideration for the node output information to the node device on the output side that outputs the node output information;
Information processing program to function as

Images