Conflict-Resilient Multi-Agent Reasoning via Signed Graph Modeling

2026-05-21 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

SIGMA, a novel SIgned Graph-informed Multi-Agent reasoning framework, addresses limitations in LLM-based multi-agent systems (MAS) that assume uniformly cooperative interactions. Existing graph-based MAS propagate errors and lack explicit modeling of conflicting inter-agent relations. SIGMA explicitly captures trust, conflict, and neutral relations via a signed relational graph. It selects relevant agents, constructs a confidence-weighted signed interaction graph, and performs conflict-aware signed message passing to reinforce trustworthy information while suppressing conflicting signals. Extensive experiments on six benchmark datasets, including MMLU, MMLU-Pro, GPQA, GSM8K, MultiArith, and HumanEval, using gpt-5.4 and gpt-5.4-mini backbones, demonstrate SIGMA consistently outperforms state-of-the-art baselines, achieving notable gains in accuracy and conflict-resilient performance, with an average accuracy of 89.17%.

Key takeaway

For AI Scientists and Machine Learning Engineers developing multi-agent LLM systems, you should consider integrating signed graph modeling to enhance system robustness. Explicitly accounting for trust and conflict among agents, rather than assuming uniform cooperation, can significantly improve accuracy and resilience against noisy or unreliable outputs. Implement conflict-aware message passing to prevent error propagation and ensure more consistent, high-quality predictions in complex reasoning tasks.

Key insights

Signed graphs explicitly model trust and conflict in multi-agent LLM systems, enhancing reasoning robustness.

Principles

• Explicitly model trust, conflict, and neutral inter-agent relations.
• Reinforce supportive signals; suppress conflicting information.
• Balance semantic relevance, diversity, and agent confidence.

Method

SIGMA performs query-guided agent selection, constructs a signed relational graph, uses conflict-aware signed message passing, and concludes with structure- and conflict-aware weighted aggregation.

In practice

• Separate positive and negative agent representations for message passing.
• Normalize signed interaction matrices for stable propagation.

Topics

• Multi-Agent Systems
• Large Language Models
• Signed Graphs
• Conflict-Resilient Reasoning
• Graph Neural Networks
• Agent Selection

Code references

• habedi/cogitator
• dj-sorry/self_consistency
• chuanyang-Zheng/Progressive-Hint
• moa-engine/MOA
• wenzhe-li/Self-MoA

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• LLMs+Graphs: Toward Graph-Native, Synergistic AI Systems — Integrating LLMs with graph computation creates synergistic AI systems that overcome LLM reasoning limitations.
• GraphDC: A Divide-and-Conquer Multi-Agent System for Scalable Graph Algorithm Reasoning — GraphDC uses a divide-and-conquer multi-agent system to scale LLM-based graph reasoning by decomposing complex graphs.
• AdaSTORM: Scaling LLM Reasoning on Dynamic Graphs via Adaptive Spatio-Temporal Multi-Agent Collaboration — AdaSTORM enables LLMs to reason on large dynamic graphs by partitioning and using spatio-temporal multi-agent collaboration.
• ARMOR-MAD: Adaptive Routing for Heterogeneous Multi-Agent Debate in Large Language Model Reasoning — Adaptive control and genuine model heterogeneity are crucial for efficient and accurate multi-agent LLM reasoning.
• LLMs+Graphs: Toward Graph-Native, Synergistic AI Systems — LLMs require graph integration to overcome structured reasoning limitations and enable graph-native AI systems.
• Trust-Aware Multi-Agent Traceability: Confidence-Calibrated Knowledge Graphs for Consistent Software Artifact Management — Calibrated confidence scores enable trust-aware coordination and consistency detection in multi-agent software engineering pipelines.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.