SFedCA: Credit Assignment-Based Active Client Selection Strategy for Spiking Federated Learning

Area of Science:

• Artificial Intelligence
• Machine Learning
• Distributed Systems

Background:

• Spiking federated learning (FL) combines privacy-preserving FL with energy-efficient spiking neural networks (SNNs).
• Current FL methods often use random client selection, ignoring data heterogeneity, which hinders model performance.
• Statistical heterogeneity in client data distribution is a key challenge in federated learning.

Purpose of the Study:

• To propose a credit assignment-based active client selection strategy for spiking federated learning (SFedCA).
• To address the limitations of random client selection in spiking FL by accounting for statistical heterogeneity.
• To enhance the convergence and precision of the global model in energy-constrained distributed learning environments.

Main Methods:

• Developed SFedCA, a client selection strategy based on credit assignment.
• Assigned client credits by analyzing the firing intensity state before and after local model training.
• Evaluated SFedCA on various non-identical and independent distribution (non-IID) scenarios.

Main Results:

• SFedCA demonstrated superior performance compared to existing state-of-the-art spiking FL methods.
• The proposed strategy requires fewer communication rounds for effective model training.
• SFedCA effectively balances the global sample distribution by judiciously selecting clients.

Conclusions:

• Credit assignment-based client selection is effective for improving spiking federated learning.
• SFedCA offers a more efficient and accurate approach to distributed learning with heterogeneous data.
• This method enhances the practical applicability of spiking FL in resource-constrained settings.