Copertina

How Does RL Post-training Induce Skill Composition? A Case Study on Countdown

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Park, Simon, Kaur, Simran, Arora, Sanjeev
Natura:	Preprint
Pubblicazione:	2025
Soggetti:	Machine Learning
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Instruct-SkillMix: A Powerful Pipeline for LLM Instruction Tuning
di: Kaur, Simran, et al.
Pubblicazione: (2024)

Can Models Learn Skill Composition from Examples?
di: Zhao, Haoyu, et al.
Pubblicazione: (2024)

Skill-Targeted Adaptive Training
di: He, Yinghui, et al.
Pubblicazione: (2025)

Prioritized Replay for RL Post-training
di: Fatemi, Mehdi
Pubblicazione: (2026)

Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR
di: Khalifa, Muhammad, et al.
Pubblicazione: (2026)

RL Grokking Recipe: How Does RL Unlock and Transfer New Algorithms in LLMs?
di: Sun, Yiyou, et al.
Pubblicazione: (2025)

Contextual Drag: How Errors in the Context Affect LLM Reasoning
di: Cheng, Yun, et al.
Pubblicazione: (2026)

On the Power of Context-Enhanced Learning in LLMs
di: Zhu, Xingyu, et al.
Pubblicazione: (2025)

Optimistic Verifiable Training by Controlling Hardware Nondeterminism
di: Srivastava, Megha, et al.
Pubblicazione: (2024)

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
di: Chu, Tianzhe, et al.
Pubblicazione: (2025)

When and How Does CLIP Enable Domain and Compositional Generalization?
di: Kempf, Elias, et al.
Pubblicazione: (2025)

Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining
di: Zhao, Rosie, et al.
Pubblicazione: (2025)

VAM: Verbalized Action Masking for Controllable Exploration in RL Post-Training -- A Chess Case Study
di: Zhang, Zhicheng, et al.
Pubblicazione: (2026)

CRePE: Convolution-aware Relative Importance in Post-training Pruning with Efficient Search
di: Park, Cheonjun
Pubblicazione: (2026)

DUMP: Automated Distribution-Level Curriculum Learning for RL-based LLM Post-training
di: Wang, Zhenting, et al.
Pubblicazione: (2025)

The unregulated plant‐based ‘milk’ industry: A threat to nutrition, health and safety?
di: Simran Kaur Arora
Pubblicazione: (2024)

On the Impossibility of Retrain Equivalence in Machine Unlearning
di: Yu, Jiatong, et al.
Pubblicazione: (2025)

On the SDEs and Scaling Rules for Adaptive Gradient Algorithms
di: Malladi, Sadhika, et al.
Pubblicazione: (2022)

SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning
di: Xia, Peng, et al.
Pubblicazione: (2026)

Generalizing from SIMPLE to HARD Visual Reasoning: Can We Mitigate Modality Imbalance in VLMs?
di: Park, Simon, et al.
Pubblicazione: (2025)

JigsawRL: Assembling RL Pipelines for Efficient LLM Post-Training
di: Hu, Zhengding, et al.
Pubblicazione: (2026)

Skill Reuse as Compression in Agentic RL
di: Xu, Zhikun, et al.
Pubblicazione: (2026)

SCALAR: Learning and Composing Skills through LLM Guided Symbolic Planning and Deep RL Grounding
di: Zabounidis, Renos, et al.
Pubblicazione: (2026)

A Quadratic Synchronization Rule for Distributed Deep Learning
di: Gu, Xinran, et al.
Pubblicazione: (2023)

SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training
di: Wang, Chen, et al.
Pubblicazione: (2025)

ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL
di: He, Zelin, et al.
Pubblicazione: (2026)

ASAP: Unsupervised Post-training with Label Distribution Shift Adaptive Learning Rate
di: Park, Heewon, et al.
Pubblicazione: (2025)

LLMStinger: Jailbreaking LLMs using RL fine-tuned LLMs
di: Jha, Piyush, et al.
Pubblicazione: (2024)

SPEED-RL: Faster Training of Reasoning Models via Online Curriculum Learning
di: Zhang, Ruiqi, et al.
Pubblicazione: (2025)

How Does Critical Batch Size Scale in Pre-training?
di: Zhang, Hanlin, et al.
Pubblicazione: (2024)

Mirage or Method? How Model-Task Alignment Induces Divergent RL Conclusions
di: Wu, Haoze, et al.
Pubblicazione: (2025)

Metacognitive Reuse: Turning Recurring LLM Reasoning Into Concise Behaviors
di: Didolkar, Aniket, et al.
Pubblicazione: (2025)

Trainable Transformer in Transformer
di: Panigrahi, Abhishek, et al.
Pubblicazione: (2023)

Provable unlearning in topic modeling and downstream tasks
di: Wei, Stanley, et al.
Pubblicazione: (2024)

Why Does RL Generalize Better Than SFT? A Data-Centric Perspective on VLM Post-Training
di: Lu, Aojun, et al.
Pubblicazione: (2026)

Unrealized Expectations: Comparing AI Methods vs Classical Algorithms for Maximum Independent Set
di: Wu, Yikai, et al.
Pubblicazione: (2025)

Learning Dynamics in RL Post-Training for Language Models
di: Tomihari, Akiyoshi
Pubblicazione: (2026)

RL in Name Only? Analyzing the Structural Assumptions in RL post-training for LLMs
di: Samineni, Soumya Rani, et al.
Pubblicazione: (2025)

Decomposing Elements of Problem Solving: What "Math" Does RL Teach?
di: Qin, Tian, et al.
Pubblicazione: (2025)

When Errors Can Be Beneficial: A Categorization of Imperfect Rewards for Policy Gradient
di: Shang, Shuning, et al.
Pubblicazione: (2026)