Fair federal learning method applied to medical field
By considering the differences in the computing resources of the hospital, using weight calculation and fairness objective function to adjust the parameter update process of federated learning, the problem of insufficient fairness in federated learning in the medical field is solved, and fairness and performance improvement in medical scenarios is achieved .
Patent Information
- Application Number
- CN202310722027.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2023-05-25
- Filing Date
- 2023-06-19
- Publication Date
- 2024-11-08
AI Technical Summary
The existing federal learning methods fail to fully consider the differences in the computing resource of hospitals in the medical field, resulting in insufficient fairness and the inability to effectively utilize the huge data sets of hospitals at all levels, and traditional methods are not suitable for imbalance in computing resource in medical scenarios.
By collecting the CPU and GPU computing resources and data volumes of each hospital, calculating the weights and randomly selecting the hospital client, using fairness objective function and fairness evaluation indicators, adjusting the parameter update process of federated learning to ensure fairness among hospitals and performance.
It realizes a federated learning method suitable for medical scenarios while ensuring patient data privacy, fairness among hospital clients and achieving certain performance.
Smart Images

Figure BDA0004291481530000031 
Figure BDA0004291481530000032 
Figure BDA0004291481530000035
Abstract
Description
Technical Field
[0001] The present invention relates to a federated learning method, and more particularly to a fair federated learning method applied in the medical field. Background Art
[0002] Artificial intelligence research in the medical field has created great value, but these studies generally require a large amount of data to support them. Hospitals at all levels have a large amount of medical data that can be used for research, but only a small number of research teams can access these data, and they can only cooperate with one or two hospitals to obtain partial data, and cannot use the huge data sets composed of hospitals at all levels. A lot of medical data in hospitals involves patient privacy. In order to protect patient privacy, the massive data of hospitals at all levels is more difficult to be used for research.
[0003] The application of federated learning technology can effectively help multiple medical institutions to integrate data and jointly model machine learning. Federated learning allows all medical data participants to jointly model through cryptographic algorithm interaction without sharing original data, breaking the data island phenomenon from a technical perspective and realizing medical big data AI collaboration.
[0004] Federated learning requires coordination between different clients to complete model training, but the data distribution between hospital clients varies greatly, which makes it difficult for the global model to meet the optimal performance of all clients and can only approach the optimal performance of the centralized training model. In addition, most federated learning work mainly uses the amount of data as the weight, but in medical scenarios, the computing resources of various hospitals vary greatly, and traditional fairness federated learning is not completely suitable for medical scenarios. Therefore, research on fairness of federated learning for medical scenarios has become very necessary.
[0005] For example, the invention patent with application number 202111381852.6 and titled Federated Learning Method, Federated Learning Device, Electronic Device, and Storage Medium discloses the existing federated learning method. However, this learning method does not take into account the computing resources of each hospital, and therefore the fairness of this learning method has certain deficiencies. Summary of the invention
[0006] In view of the deficiencies in the prior art, the object of the present invention is to provide a fair federated learning method for use in hospitals that takes into account the computing resources of the hospitals.
[0007] To achieve the above object, the present invention provides the following technical solution: a fair federated learning method applied in the medical field, comprising the following steps:
[0008] Step 1: Collect the CPU and GPU computing resources owned by hospitals at all levels to the central server through uploading by the clients of hospitals at all levels. The three attributes of data volume, CPU resources and GPU resources constitute the weight p k ;
[0009] Step 2: The central server randomly selects t hospital clients from K hospital clients and assigns them according to the weights p formed in step 1. k Calculate the effectiveness E and fairness J of the selected hospital client;
[0010] Step 3: The central server will set the server parameters ω t Distribute to the selected t hospital clients;
[0011] Step 4: The hospital client selected in step 3 updates the local parameters and
[0012] Step 5: The selected hospital client will set the local parameters and Sent to the central server, the central server updates the server parameters and obtains the server parameters ω t+1 ;
[0013] Step 6: Complete federated learning.
[0014] As a further improvement of the present invention, in step 2, after the central server randomly selects t hospital clients from K hospital clients, the effectiveness E and fairness J of the hospital clients are calculated in the following manner:
[0015] The weight p is composed of the three attributes of data volume, CPU resources and GPU resources. k And the accuracy of local data of each hospital client k To serve as utility R k , and used to evaluate the effectiveness E and fairness J of the final result, and finally obtain the following formula:
[0016] R k =a k p k
[0017]
[0018]
[0019] As a further improvement of the present invention, the local parameters updated in step 4 are and Calculated by the following formula:
[0020]
[0021]
[0022]
[0023] In the formula, ω t t is the global parameter at the moment, is the parameter of device k after multiple rounds of SGD operations on the objective function with a fixed step size at time t, is the parameter change of device k at time t, L is the learning rate, is the fairness objective function.
[0024] As a further improvement of the present invention, the fairness objective function is as follows:
[0025]
[0026] In the formula, p k is the objective function weight of fair federated learning, and q is the parameter we designed to adjust fairness.
[0027] As a further improvement of the present invention, the updating of the server parameters in step 5 is calculated and updated by the following formula:
[0028]
[0029] The beneficial effects of the present invention are as follows: by acquiring the data volume and computing power of different hospitals, the present invention combines the federated learning fairness algorithm and fairness evaluation indicators, so that the federated learning can achieve fairness among hospital clients and obtain certain performance while ensuring the privacy of hospital patient data. DETAILED DESCRIPTION
[0030] The present invention will be further described in detail with reference to the given embodiments below.
[0031] A fair federated learning method applied in the medical field in this embodiment includes the following steps:
[0032] 1) Hospital clients at all levels upload their local CPU and GPU computing resources to the central server;
[0033] 2) The central server randomly selects t clients from K hospital clients;
[0034] The weight p is composed of the three attributes of data volume, CPU resources and GPU resources. k And the accuracy of local data of each hospital client k To serve as utility R k, and is used to evaluate the effectiveness E and fairness J of the final result. The formula is as follows:
[0035] R k =a k p k
[0036]
[0037]
[0038] 3) The central server sets the parameter ω t Distribute to the selected t hospital clients;
[0039] 4) The selected hospital client updates the local parameters and This embodiment designs a new objective function based on the fairness function α-fairness. The α-fairness function is as follows:
[0040]
[0041] Among them, the fairness objective function is as follows:
[0042]
[0043] Compared with the traditional objective function, the q of the fairness objective function is similar to the design of α in α-fairness, which is a parameter used to adjust fairness. When q = 0, the objective function becomes the traditional objective function. In addition, we set p k The CPU and GPU computing resources and data volume related to the hospital are designed to replace the original data volume to reflect the high imbalance of computing resources in medical scenarios and the fairness requirements based on computing resources. The three attributes of data volume, CPU resource volume and GPU resource volume are combined to serve as the objective function weight p of fair federated learning. k Calculated based on the designed fairness objective function and The calculation formula is as follows:
[0044]
[0045]
[0046]
[0047] 5) The selected hospital client will and Sent to the central server, the central server updates the parameters and obtains ω t+1 , the calculation formula is as follows:
[0048]
[0049] 6) Complete the federated learning process across the hospital.
[0050] The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should also be regarded as the protection scope of the present invention.
Claims
1. A fair federated learning method applied in the medical field, characterized by: The steps include: Step 1: Collect the CPU and GPU computing resources owned by hospitals at all levels to the central server through uploading by the clients of hospitals at all levels. The three attributes of data volume, CPU resources and GPU resources constitute the weight p k ; Step 2: The central server randomly selects t hospital clients from K hospital clients and assigns them according to the weights p formed in step 1. k Calculate the validity E and fairness J of t hospital client data; Step 3: The central server will set the server parameters ω t Distribute to the selected t hospital clients; Step 4: The hospital client selected in step 3 updates the local parameters and Step 5: The selected hospital client will set the local parameters and Sent to the central server, the central server updates the server parameters and obtains the server parameters ω t+1 ; Step 6: Complete federated learning.
2. The fair federated learning method applied in the medical field according to claim 1, characterized in that: In step 2, after the central server randomly selects t hospital clients from K hospital clients, the effectiveness E and fairness J of the hospital clients are calculated in the following way: The weight p is composed of the three attributes of data volume, CPU resources and GPU resources. k And the accuracy of local data of each hospital client k To serve as utility R k , and used to evaluate the effectiveness E and fairness J of the final result, and finally obtain the following formula: R k =a k p k 3. The fair federated learning method applied in the medical field according to claim 1 or 2, characterized in that: The local parameters updated in step 4 and Calculated by the following formula: In the formula, ω t t is the global parameter at the moment, is the parameter of device k after multiple rounds of SGD operations on the objective function with a fixed step size at time t, is the parameter change of device k at time t, L is the learning rate, is the fairness objective function.
4. The fair federated learning method applied in the medical field according to claim 3, characterized in that: The fairness objective function is as follows: In the formula, p k is the objective function weight of fair federated learning, and q is the parameter used to adjust fairness.
5. The fair federated learning method applied in the medical field according to claim 4, characterized in that: In step 5, the server parameters are updated by calculating the following formula:
