Single-budget multi-round federated learning incentive mechanism method, device, system and medium
By modeling federated learning as a single-budget, multi-round incentive problem, and employing an approximate algorithm and a quality assessment module, the design of incentive mechanisms for non-independent, identically distributed data is solved. This achieves effective incentives and model aggregation under privacy protection, thereby improving learning performance and accuracy.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2024-03-21
- Publication Date
- 2026-06-26
AI Technical Summary
Existing technologies struggle to effectively assess the value of non-independent, identically distributed data and design incentive mechanisms to improve federated learning performance, and they also struggle to perform model aggregation while protecting privacy.
The incentive selection process is modeled as a multi-round incentive problem with a single budget. An approximate algorithm is used to evaluate individual rationality and authenticity. Participants are selected using a quality assessment module and a payoff function. A global model is constructed through an aggregation method.
It implements an effective incentive mechanism for non-independent and identically distributed data, improves the learning performance and model aggregation accuracy of federated learning, avoids model distortion caused by dishonest behavior of participants, and is suitable for privacy-preserving federated learning scenarios.
Smart Images
Figure CN118333190B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method, apparatus, system, and medium for a single-budget, multi-round federated learning incentive mechanism, belonging to the technical field of federated learning incentive mechanisms. Background Technology
[0002] Federated learning, as a framework suitable for distributed machine learning scenarios requiring privacy protection, has garnered significant attention since its inception in 2016. A well-designed incentive mechanism is crucial for the practical application of federated learning in the Internet of Things (IoT) era. This is because federated learning involves interaction between a central server and various participating clients, who are relatively independent and typically self-interested. Clients are unwilling to participate in the federated learning process without compensation. Compared to incentive mechanism designs in other fields, such as mobile crowdsourcing, cloud computing, and smart grids, federated learning presents greater challenges. Due to the inherent characteristics of federated learning, training data is distributed across different participants and is not shared, leading to unique challenges. These include data imbalance (significant differences in data volume across participants), statistical discrepancies (non-independent and identically distributed data across participants), and the potential for selfish or unreliable participating clients.
[0003] Currently, most research in the field of federated learning incentive mechanisms has not effectively addressed the following two challenges. First, due to the privacy requirements of federated learning, it is difficult to directly assess the data value and data distribution of each non-independent, identically distributed data client. Second, it is difficult to model and analyze the learning performance of federated learning within the incentive mechanism, while the performance of federated learning is undoubtedly an important aspect that must be considered in federated learning incentive mechanisms. Therefore, there is a need to provide an incentive mechanism solution for non-independent, identically distributed federated learning that considers the entire federated learning process and its performance. Summary of the Invention
[0004] In view of this, the present invention provides a method, apparatus, computer device and storage medium for a single-budget multi-round federated learning incentive mechanism. It models the incentive selection process as a single-budget multi-round incentive problem, proposes an approximate algorithm with theoretical guarantees, and ensures the rationality, authenticity and computational efficiency of individuals in the incentive process. It designs a feasible incentive mechanism for non-independent and identically distributed federated learning, which can further promote the application of federated learning.
[0005] The first objective of this invention is to provide a single-budget, multi-round federated learning incentive mechanism.
[0006] The second objective of this invention is to provide a single-budget, multi-round federated learning incentive mechanism device.
[0007] The third objective of this invention is to provide a single-budget, multi-round federated learning incentive mechanism system.
[0008] A fourth objective of this invention is to provide a computer-readable storage medium.
[0009] The first objective of this invention can be achieved by adopting the following technical solution:
[0010] The third objective of this invention can be achieved by adopting the following technical solution:
[0011] A single-budget, multi-round federated learning incentive mechanism method, the method comprising:
[0012] Receive the total budget, learning task, and standard small dataset from the task publisher;
[0013] Based on a standard small dataset, the quality of the local models of all task participants in the current round is evaluated.
[0014] Based on the selection strategy and payoff function, and combined with the quality evaluation value of each task participant's local model in the current round, the task participants participating in federated learning in the current round are selected and their remuneration is determined.
[0015] If the total budget is greater than 0 after payment, the global model is sent to each selected task participant so that each task participant can train the local model locally.
[0016] It receives the local models trained by each task participant and uses the aggregation method to perform global model aggregation.
[0017] Furthermore, the quality evaluation of the local models of all task participants in the current round is performed based on a standard small dataset, including:
[0018] Use the In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The value of each participant's contribution to the task;
[0019] make By using a preset time window and a forgetting factor to weight the historical contribution value based on freshness, we obtain the first... The quality assessment value of the local model for each task participant is determined.
[0020] Furthermore, the use of the first In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The contribution value of each participant in each task is calculated as follows:
[0021]
[0022] in, For the first The value of each participant's contribution to the task. For the first Each selected task participant takes turns The recognition accuracy, For the first The set of task participants selected to participate in federated learning. For set Any participant in the task, This represents the total number of participants in the task.
[0023] Furthermore, the method of using a preset time window and a forgetting factor to weight the historical contribution value based on freshness yields the first... The quality assessment value of each task participant's local model is given by the following formula:
[0024] ;
[0025] in, For the first Each task participant The quality assessment value of the local model, The size of the preset time window, Forgetting factor, , To contribute value to history.
[0026] Furthermore, the step of selecting task participants for the current round of federated learning and determining their remuneration based on the selection strategy and payoff function, combined with the quality evaluation value of each task participant's local model in that round, includes:
[0027] Based on the selection strategy, combined with the first Round the quality assessment value of each task participant's local model, and select the first one. The participants in each round of federated learning are as follows:
[0028] ;
[0029] in, For the first The set of task participants selected to participate in federated learning. The total number of participants in the task. The number of task participants selected in each round. For the set of combinations, from Randomly select from the task participants The set of all possible choices for each task participant. For the first The quality assessment value of the local model for each task participant. For the first Each task participant The quote;
[0030] Based on the payment function, combined with the first The quality assessment value of each task participant's local model is used to determine the first round. The compensation for participants selected to participate in the federated learning task is as follows:
[0031] ;
[0032] in, Indicates the first The remuneration for participants selected to participate in the federal learning mission. Indicates the first The round of participants selected to participate in the federated learning task. , Represents the set of combinations, excluding task participants. Other Randomly select from the task participants The task participants are a set of all possible choices.
[0033] Furthermore, the global model aggregation using the aggregation method includes:
[0034] The arithmetic mean aggregation method is used to perform an arithmetic mean of the local model weights of all selected task participants in each round, thereby completing the aggregation of the global model.
[0035] Furthermore, the global model aggregation using the aggregation method includes:
[0036] The contribution value weighted aggregation method is used to perform weighted aggregation of global model weights based on the contribution value of each selected task participant after training in each round.
[0037] The second objective of this invention can be achieved by adopting the following technical solution:
[0038] A single-budget, multi-round federated learning incentive mechanism device, characterized in that the device comprises:
[0039] The receiving module is used to receive the total budget, learning task, and standard small dataset from the task publisher.
[0040] The quality assessment module is used to assess the quality of the local models of all task participants in the current round based on a standard small dataset.
[0041] The incentive selection and reward payment module is used to select task participants participating in federated learning in the current round and determine the reward for each task participant, based on the selection strategy and payment function, combined with the quality evaluation value of the local model of each task participant in the current round.
[0042] The sending module is used to send the global model to each selected task participant if the total budget is greater than 0 after payment, so that each task participant can train the local model locally.
[0043] The model aggregation module is used to receive the local models trained by each task participant and perform global model aggregation using the aggregation method.
[0044] The third objective of this invention can be achieved by adopting the following technical solution:
[0045] A single-budget, multi-round federated learning incentive mechanism system, the system comprising a task publisher, task participants, and a cloud platform, the cloud platform being connected to the task publisher and task participants respectively;
[0046] The task publisher is used to send the total budget, learning tasks, and standard small datasets to the cloud platform, and to receive the final global model and remaining budget from the cloud platform.
[0047] The task participants are responsible for training the local model locally and sending the trained local model to the cloud platform.
[0048] The cloud platform is used to implement the single-budget, multi-round federated learning incentive mechanism described above.
[0049] The fourth objective of this invention can be achieved by adopting the following technical solution:
[0050] A computer-readable storage medium storing a program that, when executed by a processor, implements the above-described single-budget multi-round federated learning incentive mechanism method.
[0051] The present invention has the following advantages over the prior art:
[0052] 1. This invention is the first in the field of federated learning incentive mechanisms to propose a single-budget, multi-round federated learning incentive mechanism. It can be directly applied to federated learning incentive scenarios and is suitable for non-independent, identically distributed federated learning. The single-budget, multi-round incentive setting makes the incentive mechanism more convenient to apply in practice and the learning performance is easier to evaluate. On widely recognized datasets, it demonstrates significant advantages in terms of communication time and budget savings. Furthermore, it is flexible, simple, and has broad application prospects.
[0053] 2. The quality assessment module of this invention can be adapted to federated learning applications with non-independent and identically distributed data. The incentive selection and reward payment module considers the scenario setting of using a single budget to incentivize multiple rounds, which is more in line with the actual training of multiple rounds in federated learning. This makes the incentive mechanism framework more convenient in actual federated learning applications and the learning performance easier to evaluate. The model aggregation module is for privacy-preserving federated learning. It does not need to know the size of its local data to complete the aggregation of the global model, avoiding the model aggregation distortion problem caused by dishonest reporting by participating clients, and is fully compatible with the entire incentive mechanism. Attached Figure Description
[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.
[0055] Figure 1 This is an architecture diagram of the single-budget multi-round federated learning incentive mechanism system of Embodiment 1 of the present invention.
[0056] Figure 2 This is a flowchart of the single-budget multi-round federated learning incentive mechanism method of Embodiment 1 of the present invention.
[0057] Figure 3 This is a structural block diagram of the single-budget multi-round federated learning incentive mechanism device according to Embodiment 2 of the present invention. Detailed Implementation
[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0059] Example 1:
[0060] like Figure 1As shown, this embodiment provides a single-budget multi-round federated learning incentive mechanism system. The system is based on reverse auction and is oriented towards non-independent and identically distributed data. It includes a task publisher, task participants, and a cloud platform. The cloud platform is connected to the task publisher and task participants respectively. The task publisher can send the total budget, learning tasks, and standard small datasets to the cloud platform, and receive the final global model and remaining budget from the cloud platform. The task participants can train the local model locally and send the trained local model to the cloud platform.
[0061] like Figure 1 and Figure 2 As shown, this embodiment also provides a single-budget, multi-round federated learning incentive mechanism method, which is mainly implemented through a cloud platform and includes the following steps:
[0062] S201: Receive the total budget, learning task, and standard small dataset from the task publisher.
[0063] Specifically, the cloud platform receives the total budget, learning tasks, and standard small datasets sent by the task publisher.
[0064] S202. Based on the standard small dataset, evaluate the quality of the local models of all task participants in the current round.
[0065] Further, step S202 includes:
[0066] S2021, using the first In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The value of each participant's contribution to the task.
[0067] Specifically, cloud platform users Local models for recognizing standard small datasets , can obtain the first Each selected task participant takes turns Recognition accuracy , will the The wheel was selected The accuracy of identification of individual task participants Normalization is performed to determine the contribution value of unselected task participants. Then, keeping it unchanged, we can obtain the first... The value of each participant's contribution in each task round The formula is as follows:
[0068] (1);
[0069] in, For the first The value of each participant's contribution to the task. For the first Each selected task participant takes turns The recognition accuracy, For the first The set of task participants selected to participate in federated learning. For set Any participant in the task, This represents the total number of participants in the task.
[0070] S2022, make By using a preset time window and a forgetting factor to weight the historical contribution value based on freshness, we obtain the first... The quality assessment value of the local model for each task participant is determined.
[0071] Specifically, update round number Use size Preset time window and forgetting factor Value of historical contribution After applying freshness weighting, the first... Wheel quality assessment value The formula is as follows:
[0072] (2);
[0073] in, For the first Each task participant The quality assessment value of the local model, The size of the preset time window, Forgetting factor, , To contribute value to history.
[0074] S203. Based on the selection strategy and payoff function, and combined with the quality evaluation value of the local model of each task participant in the current round, select the task participants participating in federated learning in the current round and determine the remuneration for each task participant.
[0075] Furthermore, step S203 specifically includes:
[0076] S2031. Based on the selection strategy, combined with the first... Round the quality assessment value of each task participant's local model, and select the first one. Participants in the task of participating in federal learning.
[0077] Specifically, the selection strategy for task participants is as follows (3), and the selection set is determined according to this formula. The task participant node in the selected task is... Participants in the task of participating in federal learning.
[0078] (3);
[0079] in, For the first The set of task participants selected to participate in federated learning. The total number of participants in the task. The number of task participants selected in each round. For the set of combinations, from Randomly select from the task participants The set of all possible choices for each task participant. For the first Each task participant The quote.
[0080] S2032. Based on the payment function, combined with the first... The quality assessment value of each task participant's local model is used to determine the first round. Remuneration for participants selected to participate in federal learning missions.
[0081] Specifically, the payment function is given by equation (4), and the first payment function is determined based on this equation. Remuneration for participants selected to participate in federal learning missions.
[0082] (4);
[0083] in, Indicates the first The remuneration for participants selected to participate in the federal learning mission. Indicates the first The round of participants selected to participate in the federated learning task. , Represents the set of combinations, excluding task participants. Other Randomly select from the task participants The task participants are a set of all possible choices.
[0084] This embodiment is in the... After the corresponding task participants are paid their rewards, the first round... The total budget for the wheel is as follows:
[0085] (5);
[0086] If the total budget after paying the remuneration If the budget is exhausted, the entire incentive selection process ends; if If there is a budget surplus, then proceed to this round of federal learning process and move on to step S204.
[0087] S204. Send the global model to each selected task participant so that each task participant can train the local model locally.
[0088] S205. Receive the local models trained by each task participant and use the aggregation method to perform global model aggregation.
[0089] This embodiment presents two federated learning model aggregation methods adapted to this incentive mechanism framework:
[0090] 1) Use the arithmetic mean aggregation method to perform an arithmetic mean of the local model weights of all selected task participants in each round, and complete the aggregation of the global model. The formula is as follows:
[0091] (6);
[0092] in, For the first Each selected task participant takes turns Local model weights, For the first The weights of the global model.
[0093] 2) Using a contribution value-weighted aggregation method, the global model weights are weighted and aggregated based on the contribution value of each selected task participant after training in each round, as shown in the following formula:
[0094] (7);
[0095] In this embodiment, after step S205, the process returns to step S202 until... .
[0096] Finally, the efficiency of the proposed method and baseline methods (random selection method and bid-first method) were compared on MNIST, Fashion-MNIST, and CIFAR-10. First, the standard datasets were partitioned into non-independent identically distributed datasets, and then the entire incentive mechanism framework was implemented. We conducted 20 independent experiments, and when convergence reached a certain accuracy (MNIST (93%), Fashion-MNIST (80%), CIFAR-10 (60%)), the average results for the communication rounds are shown in Table 1 below.
[0097] Table 1. Average communication rounds for each algorithm
[0098]
[0099] The average remaining budget is shown in Table 2 below.
[0100] Table 2. Average Budget Remaining Results for Each Algorithm
[0101]
[0102] It should be noted that although the method operations of the above embodiments are described in a specific order, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the described steps may be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.
[0103] Example 2:
[0104] like Figure 3 As shown, this embodiment provides a single-budget, multi-round federated learning incentive mechanism device. The device includes a receiving module 301, a quality assessment module 302, an incentive selection and reward payment module 303, a sending module 304, and a model aggregation module 305. The quality assessment module 302 provides support for quality assessment of federated learning participants; the incentive selection and reward payment module 303 provides support for node selection and incentive for federated learning participants; and the model aggregation module 305 provides a method for global model aggregation. The incentive selection and reward payment module 303 is the primary component, connecting the federated learning incentive process. The quality assessment module 302 and the model aggregation module 305 provide support during the process. The specific functions of each module are as follows:
[0105] The receiving module 301 is used to receive the total budget, learning task, and standard small dataset from the task publisher;
[0106] The quality assessment module 302 is used to assess the quality of the local models of all task participants in the current round based on a standard small dataset.
[0107] The incentive selection and reward payment module 303 is used to select task participants participating in federated learning in the current round and determine the reward for each task participant based on the selection strategy and payment function, combined with the quality evaluation value of the local model of each task participant in the current round.
[0108] The sending module 304 is used to send the global model to each selected task participant if the total budget is greater than 0 after the payment of the reward, so that each task participant can train the local model locally.
[0109] The model aggregation module 305 is used to receive the local models trained by each task participant and perform global model aggregation using the aggregation method.
[0110] It should be noted that the device provided in this embodiment is only an example of the above-described division of functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above.
[0111] Example 3:
[0112] This embodiment provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the single-budget multi-round federated learning incentive mechanism method of Embodiment 1 above, as follows:
[0113] Receive the total budget, learning task, and standard small dataset from the task publisher; evaluate the quality of the local models of all task participants in the current round based on the standard small dataset; select task participants for federated learning in the current round and determine their remuneration based on the selection strategy and payoff function, combined with the quality evaluation value of each task participant's local model; if the total budget is greater than 0 after paying the remuneration, send the global model to each selected task participant so that each task participant can train their local model locally; receive the trained local models from each task participant and aggregate the global model using the aggregation method.
[0114] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.
[0115] In this embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used or combined with an instruction execution device, apparatus, or device. In this embodiment, the computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use or combined with an instruction execution device, apparatus, or device. The computer program contained on the computer-readable storage medium can be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.
[0116] The computer-readable storage medium described above can be used to write computer programs for executing this embodiment in one or more programming languages or combinations thereof. These programming languages include object-oriented programming languages—such as Java, Python, and C++—and conventional procedural programming languages—such as C or similar programming languages. The program can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0117] In summary, this invention is the first in the field of federated learning incentive mechanisms to propose a single-budget, multi-round federated learning incentive mechanism. It can be directly applied to federated learning incentive scenarios and is suitable for non-independent, identically distributed federated learning. The single-budget, multi-round incentive setting makes the incentive mechanism more convenient to apply in practice and the learning performance is easier to evaluate. On widely recognized datasets, it demonstrates significant advantages in terms of communication time and budget savings. Furthermore, it is flexible, simple, and has broad application prospects.
[0118] The above description is merely a preferred embodiment of the present invention. It will be apparent to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects. The scope of the present invention is defined by the appended claims rather than the foregoing description, and thus all variations falling within the meaning and scope of the equivalents of the claims are intended to be included within the present invention.
Claims
1. A single-budget, multi-round federated learning incentive mechanism method, characterized in that, The method includes: Receive the total budget, learning task, and standard small datasets from the task publisher, including MNIST, Fashion-MNIST, and CIFAR-10. Based on a standard small dataset, the quality of the local models of all task participants in the current round is evaluated. Based on the selection strategy, combined with the first Round the quality assessment value of each task participant's local model, and select the first one. Participants in the task of federated learning in each round; based on the payment function, combined with the first... The quality assessment value of each task participant's local model is used to determine the first round. Remuneration for participants selected to participate in the federal learning mission; If the total budget is greater than 0 after payment, the global model is sent to each selected task participant so that each task participant can train the local model locally. Receive the local models trained by each task participant and use the aggregation method to perform global model aggregation; The quality evaluation of the local models of all task participants in the current round, based on a standard small dataset, includes: Use the In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The value of each participant's contribution to the task; make By using a preset time window and a forgetting factor to weight the historical contribution value based on freshness, we obtain the first... The quality assessment value of the local model for each task participant; The use of the first In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The contribution value of each participant in each task is calculated as follows: ； in, For the first The value of each participant's contribution to the task. For the first Each selected task participant takes turns The recognition accuracy, For the first The set of task participants selected to participate in federated learning. For set Any participant in the task, The total number of participants in the task; The method of using a preset time window and a forgetting factor to weight the historical contribution value based on freshness yields the first... The quality assessment value of each task participant's local model is given by the following formula: ； in, For the first Each task participant The quality assessment value of the local model, The size of the preset time window, Forgetting factor, , To contribute value to history.
2. The single-budget, multi-round federated learning incentive mechanism method according to claim 1, characterized in that, The selection strategy, combined with the first Round the quality assessment value of each task participant's local model, and select the first one. The participants in each round of federated learning are as follows: ； in, The number of task participants selected in each round. For the set of combinations, from Randomly select from the task participants The set of all possible choices for each task participant. For the first Each task participant The quote; The above is based on the payment function, combined with the first The quality assessment value of each task participant's local model is used to determine the first round. The compensation for participants selected to participate in the federated learning task is as follows: ； in, Indicates the first The remuneration for participants selected to participate in the federal learning mission. Indicates the first The round of participants selected to participate in the federated learning task. , Represents the set of combinations, excluding task participants. Other Randomly select from the task participants The task participants are a set of all possible choices.
3. The single-budget, multi-round federated learning incentive mechanism method according to claim 1, characterized in that, The method of using aggregation to perform global model aggregation includes: The arithmetic mean aggregation method is used to perform an arithmetic mean of the local model weights of all selected task participants in each round, thereby completing the aggregation of the global model.
4. The single-budget, multi-round federated learning incentive mechanism method according to claim 1, characterized in that, The method of using aggregation to perform global model aggregation includes: The contribution value weighted aggregation method is used to perform weighted aggregation of global model weights based on the contribution value of each selected task participant after training in each round.
5. A single-budget, multi-round federated learning incentive mechanism device, characterized in that, The device includes: The receiving module is used to receive the total budget, learning task and standard small dataset from the task publisher, wherein the standard small dataset includes MNIST, Fashion-MNIST and CIFAR-10; The quality assessment module is used to assess the quality of the local models of all task participants in the current round based on a standard small dataset. The incentive selection and reward payment module is used to select strategies and combine them with the first... Round the quality assessment value of each task participant's local model, and select the first one. Participants in the task of federated learning in each round; based on the payment function, combined with the first... The quality assessment value of each task participant's local model is used to determine the first round. Remuneration for participants selected to participate in the federal learning mission; The sending module is used to send the global model to each selected task participant if the total budget is greater than 0 after payment, so that each task participant can train the local model locally. The model aggregation module is used to receive the local models trained by each task participant and perform global model aggregation using the aggregation method. The quality evaluation of the local models of all task participants in the current round, based on a standard small dataset, includes: Use the In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The value of each participant's contribution to the task; make By using a preset time window and a forgetting factor to weight the historical contribution value based on freshness, we obtain the first... The quality assessment value of the local model for each task participant; The use of the first In each round, the local model of each selected task participant identifies a standard small dataset and normalizes it to obtain the first... The contribution value of each participant in each task is calculated as follows: ； in, For the first The value of each participant's contribution to the task. For the first Each selected task participant takes turns The recognition accuracy, For the first The set of task participants selected to participate in federated learning. For set Any participant in the task, The total number of participants in the task; The method of using a preset time window and a forgetting factor to weight the historical contribution value based on freshness yields the first... The quality assessment value of each task participant's local model is given by the following formula: ； in, For the first Each task participant The quality assessment value of the local model, The size of the preset time window, Forgetting factor, , To contribute value to history.
6. A single-budget, multi-round federated learning incentive mechanism system, characterized in that, The system includes a task publisher, a task participant, and a cloud platform, with the cloud platform connected to the task publisher and the task participant respectively. The task publisher is used to send the total budget, learning tasks, and standard small datasets to the cloud platform, and to receive the final global model and remaining budget from the cloud platform. The task participants are responsible for training the local model locally and sending the trained local model to the cloud platform. The cloud platform is used to execute the single-budget multi-round federated learning incentive mechanism method as described in any one of claims 1-4.
7. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the single-budget multi-round federated learning incentive mechanism method as described in any one of claims 1-4.