AI network benefit balanced distribution system based on game theory

Abstract

The invention relates to the technical field of AI networks, discloses an AI network benefit balanced distribution system based on a game theory, and aims at designing a reputation income distribution mechanism based on node contribution degree dynamic evaluation. The mechanism can record and analyze the performance of the node in different tasks in real time or regularly, so that the contribution degree of the node is evaluated accurately, and the corresponding reputation income is distributed according to the contribution degree, and the balance consideration of the long-term contribution and the short-term performance of the node is integrated in the reputation income distribution mechanism, so that the reliability of the reputation income distribution mechanism is improved. It is ensured that rewards are based on short-term prominent performance, continuous stability and historical contribution of nodes are considered, dynamic evaluation of node contribution degree is considered, direct association of reputation income and the contribution degree is considered, balance of long-term contribution and short-term performance is considered, and the method is suitable for reward rewarding. According to the invention, the flexibility and expandability of the system are considered.

Description

Technical Field

[0001] The present invention relates to the field of AI network technology, and in particular to an AI network benefit balanced distribution system based on game theory. Background Art

[0002] The core concept of a peer-to-peer AI network is to build a decentralized, open and inclusive platform. Such a platform should allow AI developers to publish their innovative results directly and be easily used by other individuals in the network. In this process, there is no need to rely on computing power clusters monopolized by a few people to complete AI training or reasoning tasks. The traditional AI development model is often limited by large data centers or computing power clusters. These resources are usually in the hands of a few institutions or enterprises. This not only raises the threshold for AI development, but also limits the widespread dissemination and application of AI technology, making it impossible for AI to fully demonstrate its huge advantages as an advanced productivity tool in the field of innovation. The introduction of digital signature technology provides a clear definition of the ownership of data and transmission control in peer-to-peer AI networks. Through digital signatures, the development results provided by any individual for the AI ​​network can be effectively protected and confirmed, and these results can be freely integrated and used by others in the network, thereby greatly promoting the sharing and exchange of AI technology. This mechanism will make it easier for individuals to gain recognition and incentives for innovation in the field of AI, breaking the situation in the past where only a few people can obtain innovative results and enjoy the benefits they bring.

[0003] For most R&D personnel, the acquisition and use of high-performance dedicated computing clusters is not only expensive, but also faces great difficulties and obstacles. These computing clusters are often deployed in large data centers or scientific research institutions, which are difficult for ordinary individual developers to access, let alone use them to carry out cutting-edge AI model research and development. Limited by the limited resources available within a single organization, individual developers can often only conduct small-scale, low-complexity experiments within their own small scope, and cannot truly participate in the research and development of advanced large models, and thus cannot contribute to the overall development of AI technology.

[0004] Key resources such as computing power, AI models, and data are currently firmly controlled by a very small number of institutions. These institutions have continuously consolidated and expanded their position in the AI ​​market by relying on their own advantages in capital, technology, and talent, thus forming a de facto market monopoly. This monopoly situation not only hinders the normal exchange and dissemination of AI technology, but also makes it difficult for other participants in the market to obtain fair competition opportunities, thus affecting the health and vitality of the entire AI industry.

[0005] In this case, how to ensure that each participant can fairly obtain the benefits of cooperation has become an urgent problem to be solved. The concept of Shapley value in game theory provides a possible solution for this. Shapley value is a method used to measure the contribution of each participant in cooperation. It takes into account all possible participation orders and calculates the average marginal contribution of each participant in these orders. With this method, we can objectively and accurately evaluate the contribution of each participant and distribute the benefits fairly accordingly.

[0006] In AI R&D scenarios, if the concept of Shapley value can be applied to the distribution of benefits after the task is completed, it can ensure that each participant can get corresponding rewards according to their actual contribution. This can not only stimulate the enthusiasm and creativity of the participants, but also promote the widespread dissemination and in-depth application of AI technology. At the same time, by breaking the monopoly of a few institutions on resources, it lays a solid foundation for fair competition, innovation rewards and rapid development in the AI ​​industry. Now we invent an AI network benefit equilibrium distribution system based on game theory to solve the above problems. Summary of the invention

[0007] 1. Technical issues to be resolved

[0008] In view of the deficiencies in the prior art, the present invention provides an AI network benefit balance distribution system based on game theory to solve the above-mentioned problems.

[0009] (II) Technical solution

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] Preferably, the dynamic evaluation of node contribution can dynamically evaluate the service quality of the node by recording and analyzing the contribution of the node in different tasks, ensure the fairness and accuracy of reward distribution, and motivate the node to continue to provide high-quality services.

[0012] Preferably, the direct correlation between reputation benefits and contribution, by giving high-contribution nodes additional reputation benefits, encourages nodes to go all out in each task, and also promotes healthy competition and cooperation among nodes.

[0013] The optimal balance between long-term contribution and short-term performance is achieved by incorporating the node's average contribution in the past into the reward mechanism, avoiding the limitation of relying solely on short-term performance. This helps to identify and commend nodes that have provided stable and high-quality services for a long time, ensuring the continued healthy development of the system.

[0014] The system is flexible and scalable. The designed reward mechanism is highly flexible and scalable, and can be adjusted and optimized according to changes in the network environment and node requirements. This adaptability ensures that the system can maintain efficient operation in the face of complex and changing environments and continue to attract new nodes to join.

[0015] Another technical problem to be solved by the present invention is to provide an AI network benefit balance distribution system based on game theory, which is characterized by comprising the following steps:

[0016] 1) First, for each completed task, the provider (miner, proxy node or user demander) scores the computing node, and the evaluation score is quantified based on factors such as the node's computing power, task completion time, and completion quality. The evaluation score is standardized to the range of [0,1] to ensure comparability between different nodes. The proxy node can freely combine computing nodes with good credit to form a set N, while satisfying the autonomous joining and exit of the computing nodes. The computing nodes in the set are numbered: 1, 2, ..., n, that is, N = {node1, node2, ..., noden}, which is called the computing node set in the algorithm;

[0017] 2) Miners and proxy nodes: Miners and proxy nodes are responsible for task distribution and data transmission, and only receive basic income. For the income of an AI task, miners receive 20% as their basic income, and proxy nodes receive 10% as their basic income;

[0018] 3) Computing nodes: Computing nodes are responsible for executing specific computing tasks and obtaining basic income and possible reputation income. The computing nodes jointly obtain the remaining 70% of the basic income, which is distributed according to the amount and quality of tasks completed by the nodes. 2% is extracted from it and only given to computing nodes with higher reputation values, which improves the enthusiasm of nodes and reduces the malicious intentions of high-reputation nodes;

[0019] 4) The forgetting factor β(t) is used to adjust the weights of the evaluation scores at different time points to ensure that the evaluation scores of recent transactions have higher weights. For example, β(t) = e -λt , where λ is the forgetting rate.

[0020] Reputation value calculation: Calculate the value of node i at t n The reputation value R at the moment i (t n ) can be calculated by the following formula: n The evaluation score of each transaction completed within the time period can be expressed as The weighted sum of these evaluation scores can be used to obtain the computing power service at t n The reputation value R at the moment i (tn ), which can be expressed as:

[0021]

[0022] in, is at i The evaluation score at the time t0 to t n All evaluation scores within the time are accumulated;

[0023] 5) Define the value function, function v:2 N →R is for each set (a subset of N is called a union) The value function v(Z) is defined, and v(φ)=0 is stipulated.

[0024] 6) Use represents the alliance of computing nodes. Therefore, for any alliance Z belonging to N, the value function v(Z) is defined as:

[0025]

[0026] 7) In the alliance node set, the Shapley value of each computing node is expressed as:

[0027]

[0028] 8) Node Shapley Value: In a distributed network, in order to evaluate and reward the post-execution quality of data elements, an optimized Shapley value calculation formula is introduced. This formula comprehensively considers the current data execution quality of the node (recorded as "this time") and the average data execution quality in the past (recorded as "past average"), aiming to encourage nodes to continue to provide high-quality data services. The specific calculation formula is: Node Shapley Value = λ×This Data Execution Quality+μ×Past Average Data Execution Quality. Among them, λ and μ are weight coefficients, and the condition λ+v=1 is satisfied. According to the needs of different scenarios, the proportion of current and past data execution quality in the Shapley value calculation can be flexibly adjusted.

[0029]

[0030] 9) The system can set a reputation threshold R(t-min), and only nodes with a reputation value exceeding this threshold are eligible to participate in the distribution of reputation benefits. Reputation benefits are distributed according to the proportion of the node's reputation value among the nodes participating in the distribution, ensuring that high-reputation nodes receive more benefits.

[0031]

[0032] In order to motivate nodes to continue to provide high-quality services and enhance their reputation, the system will record and analyze the task contributions of nodes, and give additional rewards to nodes with continuous high reputation to encourage them to continue to provide high-quality services: record the nodes whose task contribution i>i-1 times task contribution, and distribute reputation benefits to these nodes. These outstanding nodes will be regarded as the main candidates for reputation benefit distribution and are eligible to participate in the additional reputation benefit distribution. Through this mechanism, we not only encourage nodes to go all out in each task, but also promote healthy competition and cooperation between nodes, and jointly promote the entire system to a higher quality service level.

[0033] (III) Beneficial effects

[0034] Compared with the prior art, the present invention provides an AI network benefit balance distribution system based on game theory, which has the following beneficial effects:

[0035] 1. This AI network benefit balance distribution system based on game theory can record and analyze the performance of nodes in different tasks in real time or regularly, so as to accurately evaluate their contribution and distribute corresponding reputation benefits according to the size of the contribution. By recording and analyzing the contribution of nodes in different tasks, the service quality of nodes can be dynamically evaluated, ensuring the fairness and accuracy of reward distribution, motivating nodes to continue to provide high-quality services, and encouraging nodes to go all out in each task by giving high-contribution nodes additional reputation benefits. It also promotes healthy competition and cooperation between nodes, and incorporates the consideration of the average contribution of nodes in the past into the reward mechanism, avoiding the limitation of relying solely on short-term performance. It helps to identify and commend nodes that have provided stable and high-quality services for a long time, ensuring the continued healthy development of the system. The reward mechanism designed by this patent is highly flexible and scalable, and can be adjusted and optimized according to changes in the network environment and node requirements. This adaptability ensures that the system can maintain efficient operation in the face of complex and changing environments and continue to attract new nodes to join.

[0036] 2. This AI network benefit balance distribution system based on game theory incorporates the balance consideration of the node's long-term contribution and short-term performance into the reputation benefit distribution mechanism, ensuring that rewards are not only based on outstanding short-term performance, but also take into account the node's continued stability and historical contribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the Shapley value of the calculation node under different γ and μ values ​​d of the present invention;

[0038] Figure 2 This is a graph showing the calculation results of the Shapley value in the presence of malicious nodes in the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Embodiment 1:

[0041] The AI ​​network benefit balance distribution system based on game theory is characterized by comprising the following steps:

[0042] 1) First, for each completed task, the provider (miner, proxy node or user demander) scores the computing node, and the evaluation score is quantified based on factors such as the node's computing power, task completion time, and completion quality. The evaluation score is standardized to the range of [0,1] to ensure comparability between different nodes. The proxy node can freely combine computing nodes with good credit to form a set N, while satisfying the autonomous joining and exit of the computing nodes. The computing nodes in the set are numbered: 1, 2, ..., n, that is, N = {node1, node2, ..., noden}, which is called the computing node set in the algorithm;

[0043] 2) Miners and proxy nodes: Miners and proxy nodes are responsible for task distribution and data transmission, and only receive basic income. For the income of an AI task, miners receive 20% as their basic income, and proxy nodes receive 10% as their basic income;

[0044] 3) Computing nodes: Computing nodes are responsible for executing specific computing tasks and obtaining basic income and possible reputation income. The computing nodes jointly obtain the remaining 70% of the basic income, which is distributed according to the amount and quality of tasks completed by the nodes. 2% is extracted from it and only given to computing nodes with higher reputation values, which improves the enthusiasm of nodes and reduces the malicious intentions of high-reputation nodes;

[0045] 4) The forgetting factor β(t) is used to adjust the weights of the evaluation scores at different time points to ensure that the evaluation scores of recent transactions have higher weights. For example, β(t) = e -λt , where λ is the forgetting rate.

[0046] Reputation value calculation: Calculate the value of node i at t n The reputation value R at the moment i (t n ) can be calculated by the following formula: n The evaluation score of each transaction completed within the time period can be expressed as The weighted sum of these evaluation scores can be used to obtain the computing power service at t n The reputation value R at the moment i (t n ), which can be expressed as:

[0047]

[0048] in, is at i The evaluation score at the time t0 to t n All evaluation scores within the time are accumulated;

[0049] 5) Define the value function, function v:2 N →R is for each set (a subset of N is called a union) The value function v(Z) is defined, and v(φ)=0 is stipulated.

[0050] 6) Use represents the alliance of computing nodes. Therefore, for any alliance Z belonging to N, the value function v(Z) is defined as:

[0051]

[0052] 7) In the alliance node set, the Shapley value of each computing node is expressed as:

[0053]

[0054] 8) Node Shapley Value: In a distributed network, in order to evaluate and reward the post-execution quality of data elements, an optimized Shapley value calculation formula is introduced. This formula comprehensively considers the current data execution quality of the node (recorded as "this time") and the average data execution quality in the past (recorded as "past average"), aiming to encourage nodes to continue to provide high-quality data services. The specific calculation formula is: Node Shapley Value = λ×This Data Execution Quality+μ×Past Average Data Execution Quality. Among them, λ and μ are weight coefficients, and the condition λ+μ=1 is satisfied. According to the needs of different scenarios, the proportion of current and past data execution quality in the Shapley value calculation can be flexibly adjusted.

[0055]

[0056] 9) The system can set a reputation threshold R(t-min), and only nodes with a reputation value exceeding this threshold are eligible to participate in the distribution of reputation benefits. Reputation benefits are distributed according to the proportion of the node's reputation value among the nodes participating in the distribution, ensuring that high-reputation nodes receive more benefits.

[0057]

[0058] In order to motivate nodes to continue to provide high-quality services and enhance their reputation, the system will record and analyze the task contributions of nodes, and give additional rewards to nodes with continuous high reputation to encourage them to continue to provide high-quality services: record the nodes whose task contribution i>i-1 times task contribution, and distribute reputation benefits to these nodes. These outstanding nodes will be regarded as the main candidates for reputation benefit distribution and are eligible to participate in the additional reputation benefit distribution. Through this mechanism, we not only encourage nodes to go all out in each task, but also promote healthy competition and cooperation between nodes, and jointly promote the entire system to a higher quality service level.

[0059] This patent takes into account the dynamic evaluation of node contribution. By recording and analyzing the contribution of nodes in different tasks, the service quality of nodes can be dynamically evaluated, ensuring the fairness and accuracy of reward distribution, and motivating nodes to continue to provide high-quality services. This patent considers the direct correlation between reputation benefits and contribution. This patent encourages nodes to go all out in each task by giving high-contribution nodes additional reputation benefits, and also promotes healthy competition and cooperation between nodes. This patent considers the balance between long-term contribution and short-term performance. This patent incorporates the consideration of the average contribution of nodes in the past into the reward mechanism, avoiding the limitations of relying solely on short-term performance. It helps to identify and commend nodes that have provided stable and high-quality services for a long time, ensuring the continued healthy development of the system. This patent takes into account the flexibility and scalability of the system. The reward mechanism designed by this patent is highly flexible and scalable, and can be adjusted and optimized according to changes in the network environment and node requirements. This adaptability ensures that the system can maintain efficient operation in the face of complex and changing environments and continue to attract new nodes to join.

[0060] The beneficial effects of the present invention are as follows: The core of this patent is to design a reputation benefit distribution mechanism based on the dynamic evaluation of node contribution. This mechanism can record and analyze the performance of nodes in different tasks in real time or regularly, so as to accurately evaluate their contribution and distribute corresponding reputation benefits according to the size of their contribution; this patent incorporates the balance between the long-term contribution and short-term performance of nodes into the reputation benefit distribution mechanism, ensuring that rewards are not only based on outstanding short-term performance, but also take into account the continuous stability and historical contribution of nodes. These outstanding nodes will be regarded as the main candidates for reputation benefit distribution and will be eligible to participate in additional reputation benefit distribution. Through this mechanism, we not only encourage nodes to go all out in each task, but also promote healthy competition and cooperation between nodes, and jointly promote the entire system to a higher quality service level.

[0061] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. The AI ​​network benefit balance distribution system based on game theory is characterized by: The following steps are involved: 1) First, for each completed task, the provider (miner, proxy node or user demander) scores the computing node, and the evaluation score is quantified based on factors such as the node's computing power, task completion time, and completion quality. The evaluation score is standardized to the range of [0, 1] to ensure comparability between different nodes. The proxy node can freely combine computing nodes with good credit to form a set N, while satisfying the autonomous joining and exit of the computing nodes. The computing nodes in the set are numbered: 1, 2, ..., n, that is, N = {node1, node2, ..., noden}, which is called the computing node set in the algorithm; 2) Miners and proxy nodes: Miners and proxy nodes are responsible for task distribution and data transmission, and only receive basic income. For the income of an AI task, miners receive 20% as their basic income, and proxy nodes receive 10% as their basic income; 3) Computing nodes: Computing nodes are responsible for executing specific computing tasks and obtaining basic income and possible reputation income. The computing nodes jointly obtain the remaining 70% of the basic income, which is distributed according to the amount and quality of tasks completed by the nodes. 2% is extracted from it and only given to computing nodes with higher reputation values, which improves the enthusiasm of nodes and reduces the malicious intentions of high-reputation nodes; 4) The forgetting factor β(t) is used to adjust the weights of the evaluation scores at different time points to ensure that the evaluation scores of recent transactions have higher weights. For example, β(t) = e -λt , where λ is the forgetting rate. Reputation value calculation: Calculate the value of node i at t n The reputation value R at the moment i (t n ) can be calculated by the following formula: n The evaluation score of each transaction completed within the time period can be expressed as The weighted sum of these evaluation scores can be used to obtain the computing power service at t n The reputation value R at the moment i (t n ), which can be expressed as: in, is at i The evaluation score at the time t0 to t n All evaluation scores within the time are accumulated; 5) Define the value function, function v:2 N →R is for each set (a subset of N is called a union) The value function v(Z) is defined, and v(φ)=0 is stipulated. 6) Use represents the alliance of computing nodes. Therefore, for any alliance Z belonging to N, the value function v(Z) is defined as: 7) In the alliance node set, the Shapley value of each computing node is expressed as: 8) Node Shapley Value: In a distributed network, in order to evaluate and reward the post-execution quality of data elements, an optimized Shapley value calculation formula is introduced. This formula comprehensively considers the current data execution quality of the node (recorded as "this time") and the average data execution quality in the past (recorded as "past average"), aiming to encourage nodes to continue to provide high-quality data services. The specific calculation formula is: Node Shapley Value = λ×this data execution quality+μ×past average data execution quality. Among them, λ and μ are weight coefficients, and the condition λ+μ=1 is satisfied. According to the needs of different scenarios, the proportion of current and past data execution quality in the Shapley value calculation can be flexibly adjusted. 9) The system can set a reputation threshold R(t-min), and only nodes with a reputation value exceeding this threshold are eligible to participate in the distribution of reputation benefits. Reputation benefits are distributed according to the proportion of the node's reputation value among the nodes participating in the distribution, ensuring that high-reputation nodes receive more benefits. In order to motivate nodes to continue to provide high-quality services and enhance their reputation, the system will record and analyze the task contributions of nodes, and give additional rewards to nodes with continuous high reputation to encourage them to continue to provide high-quality services: record the nodes whose task contribution i>i-1 times task contribution, and distribute reputation benefits to these nodes. These outstanding nodes will be regarded as the main candidates for reputation benefit distribution and are eligible to participate in the additional reputation benefit distribution. Through this mechanism, we not only encourage nodes to go all out in each task, but also promote healthy competition and cooperation between nodes, and jointly promote the entire system to a higher quality service level.

2. The AI ​​network benefit balance distribution system based on game theory according to claim 1 is characterized by: Dynamic evaluation of node contribution, by recording and analyzing the contribution of nodes in different tasks, can dynamically evaluate the service quality of nodes, ensure the fairness and accuracy of reward distribution, and motivate nodes to continue to provide high-quality services.

3. The AI ​​network benefit balance distribution system based on game theory according to claim 1 is characterized by: The direct correlation between reputation benefits and contribution, by giving high-contribution nodes additional reputation benefits, encourages nodes to go all out in every task, and also promotes healthy competition and cooperation among nodes.

4. The AI ​​network benefit balance distribution system based on game theory according to claim 1 is characterized by: The balance between long-term contribution and short-term performance is integrated into the reward mechanism to consider the average contribution of the node in the past, avoiding the limitation of relying solely on short-term performance. It helps to identify and commend nodes that provide stable and high-quality services for a long time, ensuring the continued healthy development of the system.

5. The AI ​​network benefit balance distribution system based on game theory according to claim 1 is characterized by: The flexibility and scalability of the system, the designed reward mechanism is highly flexible and scalable, and can be adjusted and optimized according to changes in the network environment and node requirements. This adaptability ensures that the system can maintain efficient operation in the face of complex and changing environments and continue to attract new nodes to join.

Images