Imagen de Portada

Atomic Skills are the Prerequisite: When Reinforcement Learning Synthesizes Compositional Reasoning, and When It Only Amplifies

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Cheng, Sitao, Yin, Xunjian, Zhou, Ruiwen, Li, Yuxuan, Wang, Xinyi, Pan, Liangming, Wang, William Yang, Zhong, Victor
Formato:	Preprint
Publicado:	2025
Materias:	Artificial Intelligence Computation and Language
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Understanding the Interplay between Parametric and Contextual Knowledge for Large Language Models
por: Cheng, Sitao, et al.
Publicado: (2024)

RuleArena: A Benchmark for Rule-Guided Reasoning with LLMs in Real-World Scenarios
por: Zhou, Ruiwen, et al.
Publicado: (2024)

LEDOM: Reverse Language Model
por: Yin, Xunjian, et al.
Publicado: (2025)

Differentiable Evolutionary Reinforcement Learning
por: Cheng, Sitao, et al.
Publicado: (2025)

Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement
por: Yin, Xunjian, et al.
Publicado: (2024)

Epistemic Context Learning: Building Trust the Right Way in LLM-Based Multi-Agent Systems
por: Zhou, Ruiwen, et al.
Publicado: (2026)

Ask Only When Needed: Proactive Retrieval from Memory and Skills for Experience-Driven Lifelong Agents
por: Cai, Yuxuan, et al.
Publicado: (2026)

Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts
por: Zheng, Haizhong, et al.
Publicado: (2025)

Understanding Reasoning Ability of Language Models From the Perspective of Reasoning Paths Aggregation
por: Wang, Xinyi, et al.
Publicado: (2024)

Pride and Prejudice: LLM Amplifies Self-Bias in Self-Refinement
por: Xu, Wenda, et al.
Publicado: (2024)

Self-DC: When to Reason and When to Act? Self Divide-and-Conquer for Compositional Unknown Questions
por: Wang, Hongru, et al.
Publicado: (2024)

ConciseRL: Conciseness-Guided Reinforcement Learning for Efficient Reasoning Models
por: Dumitru, Razvan-Gabriel, et al.
Publicado: (2025)

When Safe Skills Collide: Measuring Compositional Risk in Agent Skill Ecosystems
por: Wang, Su, et al.
Publicado: (2026)

COrAL: Order-Agnostic Language Modeling for Efficient Iterative Refinement
por: Xie, Yuxi, et al.
Publicado: (2024)

MARS Policy: Multimodality Only When It Matters
por: Jia, Jindou, et al.
Publicado: (2026)

The Geometry of Reasoning: Flowing Logics in Representation Space
por: Zhou, Yufa, et al.
Publicado: (2025)

Harnessing Rule-Based Reinforcement Learning for Enhanced Grammatical Error Correction
por: Li, Yilin, et al.
Publicado: (2025)

When Domains Interact: Asymmetric and Order-Sensitive Cross-Domain Effects in Reinforcement Learning for Reasoning
por: Yang, Wang, et al.
Publicado: (2026)

Do Latent-CoT Models Think Step-by-Step? A Mechanistic Study on Sequential Reasoning Tasks
por: Liang, Jia, et al.
Publicado: (2026)

Think Only When You Need with Large Hybrid-Reasoning Models
por: Jiang, Lingjie, et al.
Publicado: (2025)

Position: AI/ML Influencers Have a Place in the Academic Process
por: Weissburg, Iain Xie, et al.
Publicado: (2024)

When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning
por: Wei, Jiaqi, et al.
Publicado: (2026)

When Firing Someone Is the Only Choice
por: Turner, Anne M.
Publicado: (2004)

Debate Only When Necessary: Adaptive Multiagent Collaboration for Efficient LLM Reasoning
por: Eo, Sugyeong, et al.
Publicado: (2025)

Disentangling Memory and Reasoning Ability in Large Language Models
por: Jin, Mingyu, et al.
Publicado: (2024)

Learning When to Stop: Adaptive Latent Reasoning via Reinforcement Learning
por: Ning, Alex, et al.
Publicado: (2025)

Reason Only When Needed: Efficient Generative Reward Modeling via Model-Internal Uncertainty
por: Xue, Chao, et al.
Publicado: (2026)

SMART: When is it Actually Worth Expanding a Speculative Tree?
por: Wang, Lifu, et al.
Publicado: (2026)

When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions
por: Li, Zhening, et al.
Publicado: (2024)

Learning Shortest Paths When Data is Scarce
por: Matsypura, Dmytro, et al.
Publicado: (2026)

Towards a Mechanistic Understanding of Propositional Logical Reasoning in Large Language Models
por: Chen, Danchun, et al.
Publicado: (2026)

Partner Modelling Emerges in Recurrent Agents (But Only When It Matters)
por: Mon-Williams, Ruaridh, et al.
Publicado: (2025)

DDA-Thinker: Decoupled Dual-Atomic Reinforcement Learning for Reasoning-Driven Image Editing
por: Yang, Hanqing, et al.
Publicado: (2026)

When Autonomy Goes Rogue: Preparing for Risks of Multi-Agent Collusion in Social Systems
por: Ren, Qibing, et al.
Publicado: (2025)

Learning How to Use Tools, Not Just When: Pattern-Aware Tool-Integrated Reasoning
por: Xu, Ningning, et al.
Publicado: (2025)

When Is Compositional Reasoning Learnable from Verifiable Rewards?
por: Barzilai, Daniel, et al.
Publicado: (2026)

When Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable Rewards
por: Wang, Li, et al.
Publicado: (2026)

Call Me When Necessary: LLMs can Efficiently and Faithfully Reason over Structured Environments
por: Cheng, Sitao, et al.
Publicado: (2024)

How Is LLM Reasoning Distracted by Irrelevant Context? An Analysis Using a Controlled Benchmark
por: Yang, Minglai, et al.
Publicado: (2025)

Learning When to Translate for Multilingual Reasoning
por: Kang, Deokhyung, et al.
Publicado: (2026)