Federated K-Means Clustering With Secure Center Aggregation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The application of clustering algorithms in federated learning scenarios is lacking, particularly for k-means clustering, which hinders efficient data utilization and privacy preservation in large-scale applications.
Innovation Solution
A federated learning method for k-means clustering using secure aggregation algorithms, enabling horizontal and vertical federated learning by calculating cluster centers and labels across decentralized participants while protecting local data privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional centralized learning is used to apply clustering algorithms, then the clustering accuracy can be maintained, but the data privacy is compromised and computational burden concentrates on one device
Solution Approach 1:
The patent divides the centralized clustering computation into distributed segments across multiple participants. Each participant performs local k-means clustering on their own data to obtain cluster centers, then securely transmits only the cluster center information to a server for aggregation. This segmentation allows clustering to be performed distributedly while preserving data privacy, as raw data never leaves local devices.
Solution Approach 2:
The patent introduces a server as an intermediary that aggregates cluster centers from multiple participants without accessing their raw data. The server computes the global cluster center by aggregating encrypted or secure information from participants, acting as a mediator that enables collaborative clustering while maintaining data isolation and privacy protection.
2Object-affected harmful factors
If federated learning is applied to k-means clustering, then data privacy is preserved, but the communication efficiency and algorithm complexity increase
Solution Approach 1:
The patent extracts only the essential cluster center information from the complete k-means clustering process and transmits it to the server for aggregation. By taking out only the necessary computational results (cluster centers) rather than transmitting raw data or complete model parameters, the communication overhead is significantly reduced while maintaining both privacy and clustering effectiveness.
Solution Approach 2:
The patent changes the parameter being transmitted from raw data or complete models to simplified cluster center coordinates. This parameter transformation reduces the dimensionality and volume of communication requirements, making the federated clustering algorithm more efficient while preserving the essential clustering information needed for convergence.
3Reliability
If secure aggregation is used to protect local data, then privacy is preserved, but the communication volume increases with database size
Solution Approach 1:
The patent extracts and transmits only the aggregated cluster center information from each participant's local computation, rather than transmitting all local data or complete model parameters. This extraction approach ensures that communication volume remains constant and independent of the size of local databases, as only the final cluster center coordinates (fixed dimension) are transmitted regardless of how much local data exists.
Data Source
AI summary
Disclosed is a federated learning method for a k-means clustering algorithm. Horizontal federated learning includes the following steps: 1) initializing K clusters, and distributing, a local sample to a cluster closest to the sample; 2) calculating a new cluster center of the cluster; and 3) if the cluster center changes, then returning to step 1). Vertical federated learning includes the following steps: 1) running, the k-means clustering algorithm locally to obtain T local clusters and intersecting to obtain TL new clusters, or running an AP clustering algorithm to obtain Ti clusters and intersecting to obtain Πi=1LTi new clusters; 2) taking TL (Πi=1LTi) new cluster centers as input samples, and initializing the K clusters; 3) distributing each sample to the cluster closest to the sample; 4) calculating a new cluster center of the cluster; and 5) if the cluster center changes, then returning to step 3).