Newest papers

One Act, Many Masks: Effect-Marginal Policy Optimization for Long-Horizon Agents

Shaojin Ma, Ruitao Wang, Pengyi Li, Jie Yang, Liangyu Chen, Gangyi Zhao, Hongyao Tang, Yan Zheng, Jianye Hao · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29427 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Os avanços recentes na aprendizagem de reforço de múltiplas voltas (RL) melhoraram substancialmente o desempenho dos agentes de LLM em tarefas interativas complexas e de longo horizonte. No entanto, apesar dos progressos na atribuição de crédito de grãos finos, a otimização de políticas ainda aplica esses sinais de aprendizagem por meio de proporções de probabilidade específicas de resposta. Respostas textuais diferentes podem induzir o mesmo efeito executável, mas as diferenças de crédito decorrentes de resultados subsequentes permanecem ligadas às suas realizações textuais individuais. Identificamos esta discrepância como equilíbrio de atualização de crédito. Nossa análise revela que o acoplamento de crédito-texto forma diretamente direções de atualização de políticas, mesmo quando o contexto de política e o efeito executável

Machine translation (Meta M2M100); the English original follows.

Abstract

Recent advances in multi-turn reinforcement learning (RL) have substantially improved the performance of LLM agents on complex, long-horizon interactive tasks. However, despite progress in fine-grained credit assignment, policy optimization still applies these learning signals through response-specific probability ratios. Different textual responses can induce the same executable effect, yet credit differences arising from subsequent outcomes remain tied to their individual textual realizations. We identify this mismatch as credit-update misalignment. Our analysis reveals that credit-text pairing directly shapes policy-update directions, even when the policy context and executable effect are held fixed. Motivated by this finding, we propose Effect-Marginal Policy Optimization (EMPO), which aligns policy optimization with execution-level equivalence through an effect-level surrogate that couples updates across alternative textual realizations. EMPO preserves the original credit signals, integrates with both trajectory-level and step-level credit estimators, and requires no additional critics or rollouts. Experiments on ALFWorld, WebShop, and Sokoban show that EMPO outperforms strong baselines in both task performance and learning efficiency.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Reinforcement learning, Computer science, Executable, Context (archaeology), Task (project management), Artificial intelligence

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange