Federal learning-based noise data processing method and device, and medium

Abstract

The invention discloses a noise data processing method and device based on federated learning and a medium, relates to the field of data processing and the field of data privacy protection and machine learning optimization, and is used for solving the problem that noise intensity cannot be dynamically adjusted according to task characteristics in a model training process. The method comprises the steps of comparing a privacy consumption value of a global large model with a preset privacy budget, when the privacy consumption value exceeds the preset privacy budget, stopping updating the global large model, searching task characteristics of training tasks and noise intensity intervals of the task characteristics, and according to task characteristic priorities of the training tasks, updating the global large model; and extracting a target noise intensity interval from the noise intensity intervals, and adjusting the added noise of the training task according to the target noise intensity interval. The privacy protection intensity is evaluated through privacy budget, the noise intensity is dynamically adjusted according to task characteristics, it is ensured that the privacy of data in the training process is fully guaranteed, and sensitive information leakage is avoided.

Description

Technical Field

[0001] The present application relates to the fields of data processing, data privacy protection, and machine learning optimization, and in particular to a method, device, and medium for processing noisy data based on federated learning. Background Art

[0002] As data privacy regulations (such as GDPR) become increasingly stringent, data in various relevant areas of our lives contain a large amount of personal sensitive information. In terms of privacy protection, current centralized data processing and machine learning methods are facing increasing compliance pressure and privacy leakage risks.

[0003] Federated learning has gained wide attention as an emerging distributed machine learning method. The core idea of ​​federated learning is to achieve data privacy protection by allowing each client to train the model locally and only upload model parameters instead of original data, thus avoiding the data leakage problem in the current centralized learning method. However, in practical applications, federated learning still faces the difficulty of how to balance privacy protection and model performance, especially when the task characteristics are different. Therefore, differential privacy technology is usually complementary to federated learning technology to protect data privacy. However, it cannot dynamically adjust the noise intensity according to the specific characteristics of the task during model training, which affects the efficiency and accuracy of training. Summary of the invention

[0004] The embodiments of the present application provide a noise data processing method, device and medium based on federated learning, which are used to solve the problem that the noise intensity cannot be dynamically adjusted according to task characteristics during model training.

[0005] The present application embodiment adopts the following technical solutions: On the one hand, an embodiment of the present application provides a noise data processing method based on federated learning, the method comprising: comparing the privacy consumption value of the global large model with a preset privacy budget; when the privacy consumption value exceeds the preset privacy budget, stopping updating the global large model; finding the task characteristics of the training task and the noise intensity range of the task characteristics; each task characteristic has a corresponding noise intensity range; according to the task characteristic priority of the training task, extracting the target noise intensity range from the noise intensity range; according to the target noise intensity range, adjusting the added noise of the training task.

[0006] In one example, a target noise intensity interval is extracted from the noise intensity interval based on the task characteristic priority of the training task, specifically including: when there are multiple task characteristics, finding the overlapping interval between the noise intensity interval of the highest priority task characteristic and the noise intensity intervals of other task characteristics to obtain an overlapping interval set, and extracting the target noise intensity interval from the noise intensity interval of the highest priority task characteristic based on the overlapping interval set.

[0007] In one example, according to the overlapping interval set, the target noise intensity interval is extracted from the noise intensity interval of the highest priority task characteristic, specifically including: when the overlapping interval set is empty, determining the noise intensity interval of the highest priority task characteristic as the target noise interval.

[0008] In one example, the method also includes: when the overlapping interval set is not empty, determining whether there is a task characteristic combination with a common noise intensity interval; when there is a task characteristic combination, determining the highest priority combination in the task characteristic combination according to the task characteristic priority; and determining the common noise intensity interval of the highest priority combination as the target noise intensity interval.

[0009] In one example, the highest priority combination in a task feature combination is determined based on the task feature priority, specifically including: sorting the task features in each task feature combination in order of task feature priority from high to low; performing transitive comparison on multiple task feature combinations to obtain the highest priority combination; transitive comparison means comparing the high priority task feature combination in the comparison result with the next other task feature combination again based on the comparison result of two task feature combinations.

[0010] In one example, multiple task feature combinations are transitively compared to obtain the highest priority combination, specifically including: arbitrarily selecting two task feature combinations from multiple task feature combinations; traversing and comparing from the first sorting position, and stopping the traversal and comparison when the task features at the same sorting position in the two task feature combinations are different or do not have the same sorting position; when the task features at the same sorting position are different, determining the task feature combination with a higher task feature priority as the high priority task feature combination; or when they do not have the same sorting position, determining the task feature combination with the remaining sorting position as the high priority task feature combination; comparing the high priority task feature combination with the next other task feature combination selected again, until the next other task feature combination cannot be selected, and determining the high priority task feature combination in the last comparison result as the highest priority combination.

[0011] In one example, the method further includes: determining other task characteristics related to each overlapping interval when there is no task characteristic combination; and determining the overlapping interval of the other task characteristics with the highest priority as the target noise interval.

[0012] In one example, the method also includes: when the privacy consumption value exceeds the preset privacy budget, detecting the current model parameters of the global large model; comparing the current model parameters with the original model parameters of the global large model to obtain the model parameter change value; when the model parameter change value is greater than or equal to the preset parameter change value threshold, determining that the training task is in a privacy leakage state.

[0013] On the other hand, an embodiment of the present application provides a noise data processing device based on federated learning, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any one of the above-mentioned noise data processing methods based on federated learning.

[0014] On the other hand, an embodiment of the present application provides a non-volatile computer storage medium for noise data processing based on federated learning, which stores computer executable instructions, and the computer executable instructions can execute any of the above-mentioned noise data processing methods based on federated learning.

[0015] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: This application uses a privacy budget to evaluate the strength of privacy protection, and dynamically adjusts the noise intensity according to the characteristics of the task to ensure that the privacy of the data is fully protected during the training process, avoid sensitive information leakage, and enable the global model to minimize the negative impact of noise on the training effect while protecting privacy, thereby improving the accuracy and generalization ability of the model. At the same time, the global model can adapt to the needs of diverse tasks. Different training tasks have different requirements for privacy protection and model performance. Dynamic noise adjustment can flexibly respond to specific task characteristics and meet diverse needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solution of the present application, some embodiments of the present application will be described in detail below in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a method for processing noise data based on federated learning provided in an embodiment of the present application; Figure 2 A flowchart of dynamic adjustment of noise intensity of a noise data processing method based on federated learning provided in an embodiment of the present application; Figure 3 A privacy budget flow chart of a noisy data processing method based on federated learning provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a noise data processing device based on federated learning provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0018] Some embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0019] Figure 1 A flowchart of a method for processing noise data based on federated learning is provided in an embodiment of the present application. The method can be applied to different business fields. Certain input parameters or intermediate results in the process allow manual intervention and adjustment to help improve accuracy.

[0020] The analysis method involved in the embodiment of the present application can be implemented by a terminal device or a server, and the present application does not impose any special restrictions on this. For the convenience of understanding and description, the following embodiments are described in detail by taking a local server as an example.

[0021] Based on this, Figure 1 The process in may include the following steps: S101: Compare the privacy consumption value of the global large model with the preset privacy budget.

[0022] It should be noted that the differential privacy mechanism achieves privacy protection by adding noise to data or query results, while federated learning distributes the model training process across multiple local devices or data owners to avoid centralized transmission and storage of original data, thereby achieving privacy protection.

[0023] In the centralized data processing scenario, differential privacy requires operations such as adding noise to the entire data set. In federated learning, differential privacy technology can be applied to each data subset on each local device, and then the model can be aggregated. This can better adapt to the distributed data environment and reduce the impact of noise on model performance to a certain extent. For example: In a federated learning system composed of multiple mobile devices, each device stores user behavior data. Local data can be differentially processed on each device before local model training, and then the training results can be uploaded to the server for aggregation, thereby realizing privacy-preserving model training in a distributed environment.

[0024] The privacy budget is a measure of the degree of privacy protection of data. A smaller privacy budget value means that the algorithm has a higher degree of privacy protection because it limits the possibility of changes in the output results due to changes in a single data record. Intuitively, a smaller privacy budget value makes it more difficult for an attacker to infer information about a specific individual's data by observing the output of the algorithm; while a larger privacy budget value means that a greater privacy loss is allowed, that is, the algorithm output results may be more affected by changes in a single data record, and the degree of privacy protection is relatively low.

[0025] In some embodiments of the present application, before comparing the privacy consumption value of the global large model with the preset privacy budget, it is necessary to first analyze the task characteristics. The tasks include but are not limited to tasks in the financial field, the scientific field, the legal field, etc., and the data distribution, feature type and privacy requirements of each task are different. The present application analyzes the task characteristics through the dimensions of data sensitivity, task complexity, and training data volume. For example, some financial data (such as user personal information, account history, etc.) is more sensitive and requires stronger privacy protection measures; highly complex fraud detection tasks may require more sophisticated noise adjustment; tasks with large data volumes are less sensitive to noise and can tolerate higher intensity noise, reducing the impact on performance.

[0026] Furthermore, in order to achieve dynamic noise adjustment, we integrate the Differential Privacy (DP) mechanism into the federated learning framework. Differential privacy adds Gaussian noise when updating the model. For highly sensitive tasks, the added noise intensity is larger; for low-sensitivity tasks, the noise intensity is smaller. Different noise weighting strategies are used according to the complexity of the task and the amount of data to prevent the leakage of individual data.

[0027] Furthermore, in order to ensure the effectiveness of noise addition and the degree of privacy protection, a privacy assessment module is introduced in the training process of federated learning to regularly assess the privacy protection level after adding noise. ) to evaluate the strength of privacy protection. Specifically, by tracking the privacy budget consumption of each model update, ensure that the preset privacy budget is not exceeded during training. Each model update should ensure that the privacy budget does not exceed the preset privacy budget. = 1.

[0028] The privacy budget is used to evaluate the strength of privacy protection to ensure that the privacy of data is fully protected during the training process and to avoid the leakage of sensitive information.

[0029] S102: When the privacy consumption value exceeds the preset privacy budget, stop updating the global large model.

[0030] It should be noted that, in some embodiments of the present application, if the privacy consumption value reaches an upper limit (for example = 1), the system will stop the model update of the current global large model, and when the privacy consumption value exceeds the preset privacy budget, detect the current model parameters of the global large model, evaluate whether too much personal data has been leaked, compare the current model parameters with the original model parameters of the global large model, obtain the model parameter change value, and then set a parameter change value threshold (such as the model parameter change value is 0.1). When the model parameter change value is greater than or equal to the preset parameter change value threshold, it indicates that privacy protection is insufficient and the training task is determined to be in a privacy leakage state.

[0031] By detecting parameter changes in the global large model to determine whether privacy has been leaked, the privacy budget management and evaluation mechanism is improved to avoid the risk of privacy leakage.

[0032] S103: Find the task characteristics of the training task and the noise intensity range of the task characteristics; each task characteristic has a corresponding noise intensity range.

[0033] It should be noted that in some embodiments of the present application, specific noise intensity intervals are set according to the task characteristics of the training task, and different task characteristics correspond to different noise intensity intervals. For example: For data sensitivity characteristics, for highly sensitive data, such as user personal information, account history, transaction records, etc., the noise intensity is set to 0.3 to 0.5; for medium sensitive data, such as credit scores, loan application records, etc., the noise intensity is set to 0.15 to 0.3; for low sensitive data, such as publicly available financial statement data, the noise intensity is set to 0.05 to 0.1.

[0034] For the task complexity characteristics, high-complexity tasks, such as fraud detection, involve more complex models and a large number of features, and the noise intensity is set to 0.25 to 0.4; medium-complexity tasks, such as credit scoring, have relatively simple task models, and the noise intensity is set to 0.15 to 0.25; low-complexity tasks, such as loan default prediction, involve fewer features and simple models, and the noise intensity is set to 0.05 to 0.15.

[0035] For the data volume characteristics, for large-scale data (>100,000 data items), the noise intensity is set to 0.1 to 0.2; for medium-scale data (10,000-100,000 data items), the noise intensity is set to 0.2 to 0.3; for small-scale data (<10,000 data items), the noise intensity is set to 0.3 to 0.5.

[0036] By setting the noise intensity range for task characteristics, dynamic noise adjustment can be flexibly responded to according to specific task characteristics to meet diverse needs.

[0037] S104: Extracting a target noise intensity interval from the noise intensity interval according to the task characteristic priority of the training task.

[0038] It should be noted that before extracting the target noise intensity interval from the noise intensity interval, it is necessary to set the priority of the task characteristics, for example: data sensitivity > task complexity > data volume.

[0039] In some embodiments of the present application, when the training task has a task characteristic, the target noise intensity interval corresponding to the task characteristic of the training task is directly extracted. For example, if the task characteristic of the training task is a high-complexity task, the extracted target noise intensity interval is 0.25 to 0.4.

[0040] When a training task has multiple task characteristics, find the overlapping interval between the noise intensity interval of the highest priority task characteristic and the noise intensity interval of other task characteristics to obtain an overlapping interval set; the other task characteristics here are any task characteristics of the training task, that is, every task characteristic of the training task except the highest priority task characteristic.

[0041] When the overlapping interval set is empty, that is, the noise intensity interval of the highest priority task characteristic has no overlapping area with the noise intensity interval of any other task characteristic, then the noise intensity interval of the highest priority task characteristic is determined as the target noise interval. For example: the noise intensity interval of the highest priority task characteristic is (0, 1), and the noise intensity intervals of other task characteristics are (1, 2) and (2, 3), then the target noise interval is (0, 1).

[0042] When the overlapping interval set is not empty, determine whether there is a task characteristic combination with a common noise intensity interval; here, the task characteristic combination with a common noise intensity interval is: if the noise intensity interval of the highest priority task characteristic of the training task overlaps with the noise intensity interval of any other task characteristic of the training task, then any task characteristic that overlaps with the noise intensity interval of the highest priority task characteristic constitutes a task characteristic combination with the highest task characteristic; for example: the training task has three task characteristics A, B, and C, A is the task characteristic with the highest priority, the noise intensity interval of A overlaps with the noise intensity interval of B, then (A, B) is a task characteristic combination; the noise intensity interval of A has a common overlapping interval with the noise intensity intervals of B and C, then (A, B, C) is a task characteristic combination; more specifically: the noise intensity interval of A is (0, 6), the noise intensity interval of B is (1, 5), the noise intensity interval of C is (2, 7), and the common noise intensity interval of A, B, and C is (2, 5), so A, B, and C constitute a task characteristic combination (A, B, C).

[0043] Further, when there are task characteristic combinations, the highest priority combination in the task characteristic combinations is determined according to the task characteristic priorities, and the task characteristics in each task characteristic combination are sorted in descending order of the task characteristic priorities; Perform a transitive comparison on multiple task characteristic combinations to obtain the highest priority combination; the transitive comparison means comparing the high priority task characteristic combination in the comparison result with the next other task characteristic combination based on the comparison result of the two task characteristic combinations. The specific steps are: arbitrarily select two task characteristic combinations from multiple task characteristic combinations, and then traverse and compare the two task characteristic combinations starting from the first sorting position, and stop traversing and comparing when the task characteristics at the same sorting position in the two task characteristic combinations are different or do not have the same sorting position; when the task characteristics at the same sorting position are different, determine the task characteristic combination with a high priority as the high priority task characteristic combination; or when they do not have the same sorting position, determine the task characteristic combination with the remaining sorting position as the high priority task characteristic combination, and then compare the high priority task characteristic combination with the next other task characteristic combination selected again, until the next other task characteristic combination cannot be selected, and determine the high priority task characteristic combination in the last comparison result as the highest priority combination.

[0044] For example: There are four task characteristics A, B, C, and D, and the priority is A>B>C>D. The task characteristic combinations are (A, B, C), (A, C, D), and (A, B, C, D). Randomly select two task characteristic combinations (A, B, C) and (A, C, D), and start traversing and comparing from the first sorting position. Since the first sorting position task characteristics of the two characteristic combinations are A, the traversal continues. The priority of B is greater than the priority of C, so the characteristic combination priority size is (A, B, C) greater than (A, C, D). Then (A, B, C) and (A, B, C, D) are traversed and compared from the first sorting position. Since the first three sorting position task characteristics of the two characteristic combinations are A, B, and C, the traversal continues. Since (A, B, C) and (A, B, C, D) do not have the same sorting position in the next position, the task characteristic combination (A, B, C, D) with the remaining sorting positions is determined as the highest priority task characteristic combination.

[0045] Finally, the common noise intensity interval of the highest priority combination is determined as the target noise intensity interval.

[0046] When there is no task characteristic combination, determine other task characteristics related to each overlapping interval, and determine the overlapping interval of other task characteristics with the highest priority as the target noise interval. That is, only the task characteristic with the highest priority has an overlapping interval with a single other task characteristic. For example, if there are three task characteristics A, B, and C, and the overlapping intervals are only A and B, and A and C, then it is only necessary to compare the task characteristic priorities between B and C. Here, priority B>C, so the target noise interval is the overlapping interval of A and B.

[0047] By setting priorities for task features and selecting target noise ranges, the global model can minimize the negative impact of noise on training results while protecting privacy, thereby improving the accuracy and generalization ability of the model.

[0048] S105: Adjusting the added noise of the training task according to the target noise intensity range.

[0049] It should be noted that in some embodiments of the present application, the noise added to the training task is adjusted within the target noise intensity range, for example, using reinforcement learning to train the intelligent agent. Set the state space, action space and reward function, initialize the deep reinforcement neural network DQN, collect experience through experience replay, and use the back propagation algorithm to train the network so that the intelligent agent can adjust the noise intensity according to the real-time state of the task within the target noise intensity range, and continue to learn and optimize. After that, model training is performed. During the training process, the model will continuously adjust its own parameters through optimization algorithms (such as gradient descent, etc.) according to the characteristics and rules of local data. This parameter adjustment process is model update. For example, in a credit scoring task, the client uses the credit-related data of local customers to train the model. Parameters such as the weight and bias of the model will continue to change with training to better fit the local data and improve the accuracy of credit score prediction.

[0050] After the model training is completed, the model parameters updated after local training (usually noise is added to protect privacy) are uploaded to the server. The server collects model updates from various clients and processes these updates according to a specific aggregation strategy (such as FedAvg aggregation strategy) and integrates them into the global model.

[0051] Through this method, the global model is optimized and improved, so that the global model can integrate the information and training results of each client and improve the overall performance and generalization ability.

[0052] It should be noted that although the embodiments of the present application are based on Figure 1 Steps S101 to S105 are described in sequence, but this does not mean that steps S101 to S105 must be performed in a strict order. Figure 1 The order shown in the figure is to introduce and explain step S101 to step S105 in sequence, in order to facilitate those skilled in the art to understand the technical solution of the embodiment of the present application. In other words, in the embodiment of the present application, the order between step S101 to step S105 can be appropriately adjusted according to actual needs.

[0053] Figure 1 In the method, the privacy budget is used to evaluate the strength of privacy protection, and the noise intensity is dynamically adjusted according to the task characteristics to ensure that the privacy of the data is fully protected during the training process and avoid the leakage of sensitive information. This allows the global model to minimize the negative impact of noise on the training effect while protecting privacy, thereby improving the accuracy and generalization ability of the model. At the same time, the global model can adapt to the needs of diverse tasks. Different training tasks have different requirements for privacy protection and model performance. Dynamic noise adjustment can flexibly respond to specific task characteristics and meet diverse needs.

[0054] Figure 2 A flowchart of dynamic adjustment of noise intensity of a noise data processing method based on federated learning provided in an embodiment of the present application.

[0055] exist Figure 2 In the process, we first analyze the task characteristics. After the task characteristics analysis is completed, the differential privacy mechanism is integrated into the federated learning framework. Then, the evaluation result (whether the privacy consumption exceeds the budget) is generated through the added evaluation module. If the privacy consumption of the evaluation result exceeds the privacy budget, the noise intensity is adjusted according to the task characteristics. If the privacy consumption of the evaluation result does not exceed the privacy budget, the default noise intensity is kept unchanged. Finally, the models are combined and synchronized.

[0056] Figure 3 A privacy budget flowchart of a noisy data processing method based on federated learning provided in an embodiment of the present application.

[0057] exist Figure 3 In the process, the privacy budget is first initialized, and then the privacy consumption of the model update is observed to see whether it exceeds the budget. If the privacy consumption exceeds the budget, the budget overrun processing is performed (the noise is adjusted according to the task characteristics), and privacy leakage detection is also performed. If privacy leakage is detected, privacy leakage processing is performed (the noise is adjusted according to the task characteristics), and then the privacy protection effect is evaluated; if no privacy leakage is detected, the privacy protection effect is directly evaluated; if the privacy consumption does not exceed the budget, the model update is continued, and after the update is completed, the privacy protection effect is evaluated.

[0058] Figure 4 A structural diagram of a noise data processing device based on federated learning provided in an embodiment of the present application includes: at least one processor; and, a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, and the instructions are executed by at least one processor so that the at least one processor can perform any of the above-mentioned methods for processing noisy data based on federated learning.

[0059] Some embodiments of the present application provide a non-volatile computer storage medium for noise data processing based on federated learning, which stores computer executable instructions, and the computer executable instructions can execute any of the above-mentioned methods for noise data processing based on federated learning.

[0060] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0061] The devices and media provided in the embodiments of the present application correspond one-to-one to the methods. Therefore, the devices and media also have similar beneficial technical effects as the corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0062] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0064] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0065] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0066] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0067] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM), and non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

[0068] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0069] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0070] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the technical principle of the present application should fall within the protection scope of the present application.

Claims

1. A method for processing noisy data based on federated learning, characterized in that: The method comprises: Compare the privacy consumption value of the global large model with the preset privacy budget; When the privacy consumption value exceeds the preset privacy budget, stopping updating the global large model; Finding task characteristics of a training task and noise intensity intervals of the task characteristics; each task characteristic has a corresponding noise intensity interval; extracting a target noise intensity interval from the noise intensity interval according to the task characteristic priority of the training task; The added noise of the training task is adjusted according to the target noise intensity range.

2. The method according to claim 1, characterized in that The extracting the target noise intensity interval from the noise intensity interval according to the task characteristic priority of the training task specifically includes: When there are multiple task characteristics, finding the overlapping intervals between the noise intensity interval of the highest priority task characteristic and the noise intensity intervals of other task characteristics to obtain an overlapping interval set; According to the overlapping interval set, a target noise intensity interval is extracted from the noise intensity interval of the highest priority task characteristic.

3. The method according to claim 2, characterized in that The step of extracting a target noise intensity interval from the noise intensity interval of the highest priority task characteristic according to the overlapping interval set specifically includes: When the overlapping interval set is empty, the noise intensity interval of the highest priority task characteristic is determined as the target noise interval.

4. The method according to claim 2, characterized in that: The method further comprises: When the overlapping interval set is not empty, determining whether there is a task characteristic combination with a common noise intensity interval; When there are task characteristic combinations, determining the highest priority combination among the task characteristic combinations according to the task characteristic priorities; The common noise intensity interval of the highest priority combination is determined as the target noise intensity interval.

5. The method according to claim 4, characterized in that Determining the highest priority combination among the task characteristic combinations according to the task characteristic priorities specifically includes: Sort the task features in each task feature combination according to the order of task feature priority from high to low; A plurality of task characteristic combinations are transitively compared to obtain a highest priority combination; the transitive comparison refers to comparing the high priority task characteristic combination in the comparison result with the next other task characteristic combination again based on the comparison result of two task characteristic combinations.

6. The method according to claim 5, characterized in that The multiple task feature combinations are transitively compared to obtain the highest priority combination, specifically including: Randomly select two task characteristic combinations from multiple task characteristic combinations; Performing traversal comparison from the first sorting position, and stopping traversal comparison when the task characteristics at the same sorting position in the two task characteristic combinations are different or do not have the same sorting position; When the task characteristics at the same ranking position are different, the task characteristic combination with a higher priority is determined as the high priority task characteristic combination; or when they do not have the same ranking position, the task characteristic combination with the remaining ranking position is determined as the high priority task characteristic combination; The high-priority task feature combination is compared again with the next other task feature combination selected until the next other task feature combination cannot be selected, and the high-priority task feature combination in the last comparison result is determined as the highest priority combination.

7. The method according to claim 4, characterized in that The method further comprises: When there is no combination of task characteristics, determining other task characteristics associated with each overlapping interval; The overlapping interval of the highest priority other task characteristics is determined as the target noise interval.

8. The method according to claim 1, characterized in that The method further comprises: When the privacy consumption value exceeds the preset privacy budget, detecting current model parameters of the global large model; Comparing the current model parameters with the original model parameters of the global large model to obtain model parameter change values; When the model parameter change value is greater than or equal to a preset parameter change value threshold, it is determined that the training task is in a privacy leakage state.

9. A noise data processing device based on federated learning, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the noise data processing method based on federated learning as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium for processing noisy data based on federated learning, storing computer executable instructions, characterized in that: The computer executable instructions can execute the noise data processing method based on federated learning as described in any one of claims 1 to 8.

Images