{"slug": "learning-multi-agent-coordination-via-sheaf-admm", "title": "Learning Multi-Agent Coordination via Sheaf-ADMM", "summary": "Researchers introduced Sheaf-ADMM, a framework for multi-agent coordination that divides complex tasks into overlapping pieces assigned to individual agents. In tests, it achieved a 93% solve rate on multi-agent Sudoku versus 11% for a baseline, and retained 86% accuracy on MNIST under domain shift where a standard CNN dropped to 11%. The framework makes coordination transparent by having agents negotiate through local guesses, conflict resolution, and memory of disagreements.", "body_md": "Excited to share our paper, “Learning Multi-Agent Coordination via Sheaf-ADMM” to be presented at ICML 2026.\n\nMost AI models process information as one giant, monolithic block. But in nature, intelligence often comes from a group of individuals working together, where each individual only has a limited view of the world.\n\nWe built a framework called Sheaf-ADMM to study how this kind of collective problem-solving works. We divide a complex task into smaller overlapping pieces, and assign one agent to each piece.\n\nTo solve the global puzzle, the agents negotiate in three simple steps:\n\n-\nLocal Guesses: Every agent looks at its limited view and proposes a solution.\n\n-\nFinding Common Ground: Agents communicate with their direct neighbors to smooth out conflicts. They do not need to agree on everything, but they must agree on the boundaries where their tasks overlap.\n\n-\nRemembering Disagreements: If neighbors cannot agree, they keep a memory of that conflict. This memory forces them to try harder to compromise in the next round.\n\nWe tested this on problems where no single agent has enough information to succeed alone:\n\n-\nMulti-Agent Sudoku: Each agent sees only a single row, column, or 3x3 box. The framework achieved a 93% solve rate, while a parameter-matched message-passing baseline scored 11%.\n\n-\nImage Classification: When we tested canvas-size domain shifts, a standard CNN dropped to 11% accuracy on MNIST, while our method retained 86%.\n\n-\nMaze Pathfinding: Sheaf-ADMM matches a message-passing baseline’s accuracy while agents communicate over a 5-dimensional channel, 8x smaller than that required of the baseline (42).\n\nTraditional message-passing networks hide their reasoning inside opaque hidden states. Our framework makes coordination completely transparent. You can watch exactly how local agents debate, compromise, and eventually reach a global consensus.\n\nSheaf-ADMM draws inspiration from two fields with long histories in distributed consensus: ADMM from distributed optimization, and sheaves from applied topology. We think these perspectives may offer insights for the distributed, multi-agent AI systems increasingly being built today.", "url": "https://wpnews.pro/news/learning-multi-agent-coordination-via-sheaf-admm", "canonical_source": "https://sakana.ai/sheaf-admm/", "published_at": "2026-07-04 15:00:00+00:00", "updated_at": "2026-07-07 00:50:34.462888+00:00", "lang": "en", "topics": ["machine-learning", "ai-agents", "artificial-intelligence"], "entities": ["Sheaf-ADMM", "ICML 2026", "MNIST"], "alternates": {"html": "https://wpnews.pro/news/learning-multi-agent-coordination-via-sheaf-admm", "markdown": "https://wpnews.pro/news/learning-multi-agent-coordination-via-sheaf-admm.md", "text": "https://wpnews.pro/news/learning-multi-agent-coordination-via-sheaf-admm.txt", "jsonld": "https://wpnews.pro/news/learning-multi-agent-coordination-via-sheaf-admm.jsonld"}}