Reputation-driven Cooperation in Lattice-based Decentralized Federated Learning through Evolutionary Game Theory

2026-08-02

AI Summary

This work addresses the vulnerability of decentralized federated learning to opportunistic behaviors—such as free-riding—stemming from the absence of a central coordinator, which undermines collaboration and degrades performance. The authors propose a novel framework that integrates bounded-rational evolutionary game theory with a dynamic reputation mechanism within a lattice network topology. They formulate a utility model incorporating training costs, communication overhead, and cooperation incentives, and design a reputation-driven strategy update rule. Evaluated in peer-to-peer simulations, the approach effectively suppresses free-riding, boosting average model accuracy from 70% to 82%, elevating cooperation frequency to nearly 100% (versus less than 5% in baselines), and reducing accuracy variance from 0.40 to 0.002, thereby significantly accelerating convergence and enhancing system stability.

Abstract

Decentralized Federated Learning (DFL) has emerged as an optimal privacy-preserving solution; however, it remains vulnerable to opportunistic behaviors due to the absence of a central coordinator. While Evolutionary Game Theory (EGT) serves as a powerful framework for analyzing such behaviors, existing studies often assume that agents possess perfect rationality and maintain static strategies. To address these limitations, this paper proposes a novel EGT framework designed to analyze strategic evolution and enhance overall system performance. The primary contributions of this work are threefold: First, we model peer-to-peer (P2P) interactions on a lattice network structure under the assumption of bounded rationality. Second, we formulate a comprehensive payoff matrix incorporating training costs, communication overhead, and cooperative rewards, while tailoring a strategy update rule that captures spatial propagation dynamics. Third, we integrate a reputation-based reward-and-punishment mechanism to effectively deter free-riding behaviors. Simulation results demonstrate that the framework significantly outperforms the baseline. Specifically, it increases average accuracy from approximately 70% to 82%, elevates cooperation frequency to approach 100% (compared to below 5% in the baseline), and drops accuracy variance from around 0.40 to 0.002, thereby accelerating uniform convergence and ensuring system stability.

Problem

• Decentralized Federated Learning
• Opportunistic Behavior
• Free-riding
• Evolutionary Game Theory
• Bounded Rationality

Innovation

• Decentralized Federated Learning
• Evolutionary Game Theory
• Reputation Mechanism
• Lattice Network
• Bounded Rationality