Advocating for the Silent: Enhancing Federated Generalization for Nonparticipating Clients.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Distributed Systems

Background:

• Federated learning (FL) enables collaborative model training without sharing raw data.
• Non-IID data distributions across clients present a major challenge to FL generalization.
• Assessing model generalization is complicated by non-participating clients in FL.

Purpose of the Study:

• To address the overlooked generalization gap between participating and non-participating clients in FL.
• To develop an information-theoretic framework for quantifying FL generalization errors.
• To propose novel methods for enhancing FL model generalization on diverse and non-participating client data.

Main Methods:

• Developed an information-theoretic generalization framework for FL.
• Quantified generalization errors using information entropy and distribution discrepancies.
• Introduced a weighted aggregation approach and two client selection strategies.

Main Results:

• The proposed framework provides theoretical generalization bounds for FL.
• Weighted aggregation and client selection strategies improve generalization performance.
• Empirical evaluations validate the effectiveness of the proposed methods.

Conclusions:

• The novel framework and strategies effectively enhance FL generalization, particularly for non-participating clients.
• Addressing data distribution discrepancies is key to robust FL.
• The methods align with theoretical predictions, demonstrating practical utility.