Methods and systems for decentralized federated learning
By using decentralized federated learning, each client independently trains its local model and exchanges parameters with neighboring clients, solving the problems of high data transmission costs and high privacy risks in traditional machine learning, and achieving a system with lower latency and higher robustness.
CN116745780BActive Publication Date: 2026-07-10HUAWEI CLOUD COMPUTING TECHNOLOGIES CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAWEI CLOUD COMPUTING TECHNOLOGIES CO LTD
- Filing Date
- 2021-02-06
- Publication Date
- 2026-07-10
AI Technical Summary
Technical Problem
In traditional machine learning processes, data transmission costs are high, privacy risks are significant, and system robustness depends on centralized data centers, leading to a high risk of system failure.
Method used
By employing a decentralized federated learning approach, each client independently trains its local machine learning model. By directly exchanging model parameters and weighting coefficients with neighboring clients, communication latency is reduced, data privacy is maintained, and the system's dependence on centralized nodes is decreased.
Benefits of technology
It achieves the goals of reducing communication latency and system failure risk, balancing workload, and improving system robustness while maintaining data privacy.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116745780B_ABST
Abstract
Methods and systems for decentralized federated learning are described. Each client participating in local machine learning model training identifies one or more neighboring clients that directly communicate with the client itself. Each client sends to its neighboring clients weighted coefficients of a local model and a local model parameter set. Each client also receives from its neighboring clients respective local model parameter sets and respective weighted coefficients. Each client updates its own local model parameter set using a weighted aggregation of the received local model parameter sets, each received local model parameter set being weighted using the respective received weighted coefficient. Each client trains its local machine learning model using a machine learning algorithm and its own local data set.
Need to check novelty before this filing date? Find Prior Art