{"paper":{"title":"NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Multi-agent language systems modeled as neural networks allow reinforcement learning to induce specialization and coordination among role-free agents.","cross_cats":["cs.MA","stat.ME","stat.ML"],"primary_cat":"cs.AI","authors_text":"Haoran Lu, Luyang Fang, Ping Ma, Wenxuan Zhong","submitted_at":"2026-05-16T02:11:34Z","abstract_excerpt":"Multi-agent language systems are often built as hand-designed workflows, where agents are assigned semantic roles and communication protocols are specified in advance. We propose NeuroMAS, a method that first treats a multi-agent language system as a trainable and scalable neural-network-like architecture with LLM agents as nodes and intermediate textual signals as edges. In NeuroMAS, agent nodes are role-free but structure-aware: the topology only determines how information can flow in general, while reinforcement learning training determines how nodes communicate, specialize, and coordinate."},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show that NeuroMAS improves significantly over both inference-time and trained multi-agent baselines. We further find that organizational scaling is path-dependent: larger systems can be challenging to train from scratch, but become feasible when grown progressively from smaller trained systems.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That reinforcement learning training on the network topology can reliably induce effective specialization, communication protocols, and coordination among role-free agents without additional hand-designed constraints or semantic role assignments.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"NeuroMAS reframes multi-agent language systems as neural architectures where LLM agents learn coordination via reinforcement learning rather than predefined roles.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Multi-agent language systems modeled as neural networks allow reinforcement learning to induce specialization and coordination among role-free agents.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a306712a24135de6daf1b6d09fb6db3b76609cb35cc86c1c0a43857ca014b446"},"source":{"id":"2605.16757","kind":"arxiv","version":1},"verdict":{"id":"d120e690-3c72-4fab-86f5-9e3120b9b84a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T21:13:59.760581Z","strongest_claim":"Experiments show that NeuroMAS improves significantly over both inference-time and trained multi-agent baselines. We further find that organizational scaling is path-dependent: larger systems can be challenging to train from scratch, but become feasible when grown progressively from smaller trained systems.","one_line_summary":"NeuroMAS reframes multi-agent language systems as neural architectures where LLM agents learn coordination via reinforcement learning rather than predefined roles.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That reinforcement learning training on the network topology can reliably induce effective specialization, communication protocols, and coordination among role-free agents without additional hand-designed constraints or semantic role assignments.","pith_extraction_headline":"Multi-agent language systems modeled as neural networks allow reinforcement learning to induce specialization and coordination among role-free agents."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.16757/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-19T21:31:19.372595Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T21:21:24.005595Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T19:01:56.321919Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T18:33:26.453462Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"463c3e8127d9bb22da1622119e3f3e290fc251aa623b4f29b79f03bc06930d47"},"references":{"count":65,"sample":[{"doi":"","year":1992,"title":"Machine Learning , volume =","work_id":"3f4206f6-1e77-48fd-b656-cc7c4a027789","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"Parameter-Efficient Transfer Learning for","work_id":"ec25a416-e4a9-47ba-9a72-b1c99c11efbf","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle =","work_id":"55c0e00e-cd66-4668-b59e-52cd4c7986ab","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Learning Decentralized","work_id":"30535028-eda1-4dc0-9f74-7c2e71802c10","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Yang, An and Li, Anfeng and Yang, Baosong and Zhang, Beichen and Hui, Binyuan and Zheng, Bo and others , journal =","work_id":"8582e112-cf97-4829-b163-eafc030129a4","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":65,"snapshot_sha256":"06dc5df4f487d1c5673d1f44b4a2742b30917b6c177e0fc7430532eda9db7667","internal_anchors":7},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}