Federal learning method and system based on hierarchical clustering

A learning method and hierarchical clustering technology, applied in the fields of instruments, character and pattern recognition, computer components, etc., can solve the problem of unsatisfactory resistance to malicious attacks, and achieve the effect of resisting attacks and reducing the impact.

Active Publication Date: 2021-12-10
SUN YAT SEN UNIV
View PDF4 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the defects of the above-mentioned prior art that the anti-malicious attack effect is not ideal under the non-independent and identical distribution of data, the present invention provides a federated learning method and system based on hierarchical clustering

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Federal learning method and system based on hierarchical clustering
  • Federal learning method and system based on hierarchical clustering
  • Federal learning method and system based on hierarchical clustering

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0074] This embodiment proposes a federated learning method based on hierarchical clustering, refer to Figure 1-4 , including the following steps:

[0075] S1: Clustering the client: According to the local model uploaded by the client received by the server, the client is clustered, and the client is divided into k categories, where k is a preset positive integer; the specific steps include:

[0076] S1.1: Calculate the Euclidean distance between local models uploaded by s clients to obtain the corresponding distance matrix D;

[0077] S1.2: For each client, select the closest client to the mth client according to the distance matrix D clients as a neighbor set Neighborhood(m); where m=1, 2, . . . , k;

[0078] S1.3: Calculate the similarity between every two clients among the s clients, and obtain the similarity matrix Sim, where the similarity between every two clients refers to the division of the number of intersections of the neighbors of the two clients by which i...

Embodiment 2

[0115] This embodiment proposes a hierarchical clustering-based federated learning system, which is applied to the hierarchical clustering-based federated learning method proposed in Embodiment 1. like Figure 5 Shown is a schematic diagram of the hierarchical clustering-based federated learning system of this embodiment.

[0116] A federated learning system based on hierarchical clustering proposed in this embodiment includes:

[0117] The clustering module is used to perform clustering processing on the client according to the partial model uploaded by the client received by the server, and divide the client into k classes; the client class is transmitted to the selection module;

[0118] The selection module is used to select representatives of k clusters in each cluster; the selected representatives are transmitted to the calculation weight module;

[0119] The global model update module is used to recalculate the weight of the representatives selected by the selection m...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides a federated learning method and system based on hierarchical clustering in order to overcome the defect that the hostile attack resisting effect is not ideal under the condition that federated learning data are not independently and identically distributed. The method comprises the following steps: carrying out clustering processing on clients according to local models uploaded by the clients and received by a server, and dividing the clients into k classes; in each class, respectively selecting representatives of k client classes by adopting an aggregation algorithm of traditional distributed machine learning; recalculating the weights of the selected representatives, weakening the influence of malicious representatives possibly existing, and carrying out weighted averaging on the selected representatives to obtain an updated global model. Therefore, the attack of a malicious client is effectively resisted under the condition that the federal learning data is not independently and identically distributed.

Description

technical field [0001] The present invention relates to the field of federated learning, and more particularly, to a federated learning method and system based on hierarchical clustering. Background technique [0002] Federated learning is a new distributed machine learning framework. Compared with traditional machine learning, federated learning pays more attention to the privacy and security of data. During the training process, the data of the client cannot leave the local, and the data of different clients cannot be shared. They can only jointly train a global model by passing parameters with the cooperation of the central server. But because the data cannot be shared and the model is trained on the client side. The server cannot directly review the client's data and models for tampering. In addition, in the federated environment, the data is not independent and identically distributed. All these make malicious attacks in the federated learning environment extremely ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & AuthorityApplications(China)
IPC IPC(8): G06K9/62
CPCG06F18/232Y02D30/70
Inventor吴维刚杨晨罡
OwnerSUN YAT SEN UNIV