Towards Fairness-Aware Federated Learning.

Area of Science:

• Machine Learning
• Distributed Systems
• Artificial Intelligence

Background:

• Federated learning (FL) enables collaborative machine learning with privacy guarantees.
• Current FL models often prioritize central controller interests over client fairness.
• Unfairness in FL can reduce client participation and harm the ecosystem's sustainability.

Purpose of the Study:

• To provide a comprehensive survey of fairness-aware federated learning (FAFL).
• To analyze existing FAFL approaches by examining their assumptions and fairness notions.
• To establish a taxonomy of FAFL methods across key FL stages.

Main Methods:

• Systematic review of existing literature on fairness in FL.
• Analysis of fundamental assumptions and fairness definitions in FAFL.
• Development of a taxonomy categorizing FAFL approaches by FL process steps.

Main Results:

• Identified diverse FAFL approaches addressing fairness from multiple perspectives.
• Proposed a taxonomy covering client selection, optimization, contribution evaluation, and incentive distribution.
• Discussed key metrics for evaluating FAFL performance.

Conclusions:

• FAFL is crucial for ensuring equitable participation and sustainability in federated learning.
• The proposed taxonomy offers a structured overview of the FAFL landscape.
• Further research directions are identified to advance the field of FAFL.