Federal learning method and system for alleviating flow scheduling data node drift problem
By standardizing and verifying the local model in federated learning, the model with the best effect is selected for weighted aggregation, the node drift problem is solved and the effect of federated learning is improved.
Patent Information
- Application Number
- CN202411968906.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-05-13
AI Technical Summary
In federated learning, due to the differences in local data and execution processes of each node, it is easy to cause node drift problems, resulting in inconsistent local training goals between each node, reducing the effect of federated learning.
By sending global models to all nodes, the local models generated by each node based on local data are obtained, and each local model is standardized to eliminate the difference in numerical size. Then, the standardized model is verified using the verification data set, the model with the best effect is filtered out, weighted aggregation is performed, and a new global model is generated until the effect needs are met.
Effectively reduce the differences between different local models, eliminate node drift, make federated learning more focused on the established goals, and improve the effectiveness of federated learning.
Smart Images
Figure CN119990361A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of federated learning technology, and in particular to a federated learning method and system for alleviating the problem of node drift of flow modulation data. Background Art
[0002] In federated learning, each node performs local training independently. However, the local data and execution processes of different nodes are different, which makes node drift prone. This leads to inconsistent local training goals between nodes and reduces the effectiveness of federated learning. Summary of the invention
[0003] In view of the above-mentioned shortcomings of the existing technology, a federated learning method and system are provided to alleviate the problem of node drift of flow adjustment data, which can effectively reduce the differences between different local models, eliminate node drift, and allow federated learning to focus more on the established goals.
[0004] To achieve the above objectives, the technical solution of the present invention includes the following contents.
[0005] A federated learning method for alleviating the problem of node drift of flow adjustment data is applied to a server, and the method includes:
[0006] Send the global model g to all nodes t , to obtain the local model generated by each node based on local data Where t is the training round of federated learning, and i is the node number;
[0007] For each local model Perform standardization operations to obtain a standardized model
[0008] Use the validation dataset to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1 ;
[0009] Use the validation dataset to test the global model g t+1 Test, and when the global model g t+1 When the effect meets the requirements, the federated learning training is ended.
[0010] Furthermore, the standardized model Among them, ||·|| means finding the norm.
[0011] Furthermore, the validation data set is used to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1,include:
[0012] Use the validation dataset to standardize each model Verify and obtain the standardized model Verification indicators
[0013] Based on verification indicators For all standardized models Sort and retain the m standardized models with the best effect
[0014] According to the verification index Computational Standardization Model The weight p i ;
[0015] Using weight p i The m standardized models with the best effect Perform weighted aggregation to obtain the global model g t+1 .
[0016] Furthermore, the weight
[0017] Furthermore, the global model g is tested using the validation data set. t+1 After the inspection, it also includes:
[0018] When the global model g t+1 If the effect of g does not meet the requirements, set t = t + 1 and re-execute the sending of the global model g to all nodes. t .
[0019] A federated learning system for alleviating the problem of node drift of flow adjustment data, the system comprising:
[0020] Servers for:
[0021] Send the global model g to all nodes t , to obtain the local model generated by each node based on local data Where t is the training round of federated learning, and i is the node number;
[0022] For each local model Perform standardization operations to obtain a standardized model
[0023] Use the validation dataset to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1 ;
[0024] Use the validation dataset to test the global model g t+1 Test, and when the global model g t+1 When the effect meets the requirements, the federated learning training ends;
[0025] Node, used for the global model g t , local model generated based on local data
[0026] An electronic device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, the processor implements any of the above-mentioned federated learning methods for alleviating the problem of node drift of flow adjustment data.
[0027] A computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement any of the above-mentioned federated learning methods for alleviating the problem of node drift of flow scheduling data.
[0028] Compared with the existing technology, the present invention can eliminate the differences in numerical values between local models, retain only the differences in the direction of parameter vectors, and then screen according to a unique standard, and only aggregate local models that better meet the training objectives. This can effectively reduce the differences between different local models, eliminate node drift, and allow federated learning to focus more on established goals. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 Flowchart of the federated learning method for alleviating the problem of node drift in flow survey data. DETAILED DESCRIPTION
[0030] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.
[0031] The present invention runs on the server side, firstly standardizes the local models uploaded by the nodes to eliminate the differences in the numerical values of the local models; then, the method verifies the effects of all local models through a verification data set, and screens the models according to the performance of the local models, eliminating some models with poor performance; then, the screened local models are weighted and aggregated according to their performance on the verification data set to obtain a new global model.
[0032] like Figure 1 The federated learning method of the present invention includes the following steps 1 to 2.
[0033] Step 1: Identify the two roles of federated learning, the server and the node. The server prepares the global model, local training data requirements, and the validation data set, where the validation data set is used to verify the effects of all models. The node prepares the corresponding data according to the local training data requirements.
[0034] Step 2: The server sends the global model g to all nodes t , allowing nodes to train nodes locally using local data.
[0035] Step 3: The node performs local training on the model sent by the server to obtain a local model Send the local model to the server.
[0036] Step 4: The server performs standardization operations on each local model. The specific process is as follows:
[0037]
[0038] Here, ||·|| indicates obtaining a norm, which may be any norm, including but not limited to: 1-norm, 2-norm, ∞-norm, etc.
[0039] Step 5: Use the validation dataset to test each Verify and get verification indicators according to For all Sort and retain the first m models. Among them, the verification index It can be any indicator that can quantitatively reflect the model effect, including but not limited to: accuracy, recall, F1 value, etc.; m can be any integer greater than 0. However, when m=1, the local model with the best effect on the validation dataset will be used as the global model.
[0040] Step 6: According to Calculate p i , the specific process is:
[0041]
[0042] Step 7: Using p i Perform weighted aggregation on the m local models that pass the screening to obtain a new global model g t+1 , the specific process is as follows:
[0043]
[0044] Step 8: Use the validation dataset to test g t+1 Test, if the effect does not meet the requirements, then return to step 2; if g t+1 If the requirements can be met, then federated learning is terminated.
[0045] To sum up, in the face of the problem of large differences between the local model and the global model targets, the present invention can effectively highlight the differences between the global model and the local model through the standardized operation of the paradigm operation, so as to further better adjust and discard the local model with large differences from the global model target by setting the threshold m.
[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the claims of the present invention.
Claims
1. A federated learning method for alleviating the problem of node drift of flow adjustment data, characterized in that: Applied to a server, the method comprises: Send the global model g to all nodes t , to obtain the local model generated by each node based on local data Where t is the training round of federated learning, and i is the node number; For each local model Perform standardization operations to obtain a standardized model Use the validation dataset to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1 ; Use the validation dataset to test the global model g t+1 Test, and when the global model g t+1 When the effect meets the requirements, the federated learning training is ended.
2. The method according to claim 1, characterized in that The standardized model Among them, ||·|| means finding the norm.
3. The method according to claim 1, characterized in that The validation dataset is used to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1 ,include: Use the validation dataset to standardize each model Verify and obtain the standardized model Verification indicators Based on verification indicators For all standardized models Sort and retain the m standardized models with the best effect According to the verification index Computational Standardization Model The weight p i ; Using weight p i The m standardized models with the best effect Perform weighted aggregation to obtain the global model g t+1 .
4. The method according to claim 3, characterized in that The weight 5. The method according to claim 1, characterized in that The global model g is tested using the validation dataset. t+1 After the inspection, it also includes: When the global model g t+1 If the effect of g does not meet the requirements, set t = t + 1 and re-execute the sending of the global model g to all nodes. t .
6. A federated learning system for alleviating the problem of node drift of flow dispatching data, characterized in that: The system comprises: Servers for: Send the global model g to all nodes t , to obtain the local model generated by each node based on local data Where t is the training round of federated learning, and i is the node number; For each local model Perform standardization operations to obtain a standardized model Use the validation dataset to standardize the model Validate and use the m standardized models with the best results Generate a global model g t+1 ; Use the validation dataset to test the global model g t+1 Test, and when the global model g t+1 When the effect meets the requirements, the federated learning training ends; Node, used for the global model g t , local model generated based on local data 7. An electronic device, characterized in that: The electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the federated learning method for alleviating the problem of node drift of flow adjustment data as described in any one of claims 1-5.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement a federated learning method for alleviating the problem of node drift of flow scheduling data as described in any one of claims 1-5.