Federated multi agent system improves collaboration and privacy over rounds
PFedMA: Who Should Join Next as Federated Multi-Agent Orchestration Evolves?
Artificial Intelligence
Summary
When multiple AI agents work together without sharing raw data, it’s hard to pick the best agents each time. The authors propose PFedMA, a system that remembers each agent’s past performance and privacy settings to choose who joins next. It also lets clients update their privacy rules locally while learning from others, keeping their data safe and personalized. Their tests show PFedMA helps the agents work better together, keeps privacy risks lower, and makes the system more reliable.
What this means in practice
- •For data privacy teams: Manage multi-agent collaborations that respect diverse and evolving privacy rules across many clients without sharing raw data.
- •For ai platform engineers: Build federated AI services that adaptively select reliable agents and improve reasoning by incorporating feedback over multiple rounds.
Authors
Shuai Li, Shengyuan Liu, Fan Qi, Xiaoshan Yang
Abstract
Federated Multi-Agent Systems (FedMAS) offer a promising paradigm for collaborative reasoning among LLM-based agents while keeping raw client data decentralized. However, reliable multi-round collaboration remains challenging. Agent selection must take into consideration evolving task requirements and the agent's historical reliability, while privacy rules must adapt to changing collaboration contexts and meet the client's personalized needs. To address these challenges, we propose PFedMA, a federated multi-agent framework for feedback driven multi-round collaboration under heterogeneous privacy requirements. Across rounds, PFedMA maintains historical capability and reliability states to guide agent selection, evaluates task responses before aggregation, and refines subsequent queries according to accumulated collaboration feedback. PFedMA enables bidirectional local to global and global to local private rule evolution across heterogeneous clients. Clients generalize locally updated privacy rules and selectively incorporate globally maintained rules, thereby facilitating cross-client privacy rule evolution while preserving personalized constraints. Extensive experiments across reasoning, privacy sensitive, and adversarial settings show that PFedMA improves task performance, reduces measured privacy leakage, and improves system robustness.