[2603.27977] SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology

Computer Science > Artificial Intelligence arXiv:2603.27977 (cs) [Submitted on 30 Mar 2026] Title:SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology Authors:Yifan Wang, Bolian Li, David Cho, Ruqi Zhang, Fanping Sui, Ananth Grama View a PDF of the paper titled SARL: Label-Free Reinforcement Learning by Rewarding Reasoning Topology, by Yifan Wang and 5 other authors View PDF HTML (experimental) Abstract:Reinforcement learning has become central to improving large reasoning models, but its success still relies heavily on verifiable rewards or labeled supervision. This limits its applicability to open ended domains where correctness is ambiguous and cannot be verified. Moreover, reasoning trajectories remain largely unconstrained, and optimization towards final answer can favor early exploitation over generalization. In this work, we ask whether general reasoning ability can be improved by teaching models how to think (the structure of reasoning) rather than what to produce (the outcome of reasoning) and extend traditional RLVR to open ended settings. We introduce structure aware reinforcement learning (SARL), a label free framework that constructs a per response Reasoning Map from intermediate thinking steps and rewards its small world topology, inspired by complex networks and the functional organization of the human brain. SARL encourages reasoning trajectories that are both locally coherent and globally efficient, shifting supervision from destination ...