Research map / Multi-agent systems and games / LLM agents
HiveMind: Contribution-Guided Online Prompt Optimization of LLM Multi-Agent Systems
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes HiveMind, which uses DAG-Shapley (Shapley attribution with coalition pruning and memoization) to find underperforming agents in a 7-agent LLM stock-trading system and refines their prompts online via LLM reflection (CG-OPO).
Problem
Evaluating each agent's contribution in LLM multi-agent systems and optimizing underperforming agents online remain open; exact Shapley attribution is exponentially expensive with LLM calls.
Approach
Use Shapley values (with Sharpe-ratio coalition value) to identify the bottleneck agent in a DAG-structured agent workflow, prune non-viable coalitions and cache intermediate agent outputs (DAG-Shapley), then rewrite that agent's prompt from reflection on its failure and success cases (CG-OPO).
Key results
- CG-OPO returned 13.72% on META in the bull market vs 4.44% for w/o CG-OPO.
- DAG-Shapley reduces LLM calls from 448 to 73 (83.7% reduction) versus classic Shapley.
- CG-OPO achieved +15.15% on META and +12.62% on MSFT during the bear market, while Buy & Hold lost 3.75% and 9.50%.
Surprising findings
- NVDA bear-market return under CG-OPO was -21.81%, worse than Buy & Hold's -20.71% in Sharpe terms and higher drawdown (38.11% vs 28.41%).
Limitations
- Evaluated only on four technology stocks over a 3-month period in two market regimes.
- CG-OPO shows higher volatility and maximum drawdowns than some baselines.
Experimental setup
Baselines: w/o CG-OPO MACD SMA Buy & Hold Classic Shapley
Builds on: Shapley value verbal reinforcement learning
Closest papers by meaning
- AgentNet: Decentralized Evolutionary Coordination for LLM-based Multi-Agent Systems (NeurIPS 2025 · similarity 0.87)
- Trade in Minutes! Rationality-Driven Agentic System for Quantitative Financial Trading (ICLR 2026 · similarity 0.87)
- Shapley-Coop: Credit Assignment for Emergent Cooperation in Self-Interested LLM Agents (NeurIPS 2025 · similarity 0.87)
- R&D-Agent-Quant: A Multi-Agent Framework for Data-Centric Factors and Model Joint Optimization (NeurIPS 2025 · similarity 0.86)
- FinCon: A Synthesized LLM Multi-Agent System with Conceptual Verbal Reinforcement for Enhanced Financial Decision Making (NeurIPS 2024 · similarity 0.86)
- Stochastic Self-Organization in Multi-Agent Systems (ICLR 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.