arXiv subjects

Zhaochun Li

Publications and source records attributed to Zhaochun Li.

5 recordsLinked to original sources

arXiv2026

Distilled Reinforcement Learning for LLM Post-training

Large language model (LLM) post-training is essential for improving reasoning, adaptation, and alignment. Existing methods mainly follow two paradigms: reinforcement learning (RL) and on-policy distillation (OPD). However, RL relies on coarse-grained outcome supervision, resulting in difficult credit assignment and limited capability to acquire new knowledge. OPD, meanwhile, unconditionally matches teacher logits through KL divergence, which creates a dilemma: similar teachers provide little new knowledge, while substantially different teachers often yield ineffective guidance, largely restricting OPD to within-family distillation. We propose Distilled Reinforcement Learning (Distilled RL), which integrates teacher supervision into the RL objective to provide fine-grained guidance, selectively transfer new knowledge and avoid unconditional imitation. Distilled RL contains three components: reverse importance sampling with clipping, negative sample reset, and sequence-level geometric normalization. Through a concise and interpretable case study, we demonstrate that Distilled RL can effectively transfer previously unavailable knowledge from a teacher model to a student model. Extensive experiments across both within-family and cross-family distillation settings show that Distilled RL substantially outperforms standard RL and OPD in terms of both pass@1 and pass@k. Our code is available at https://github.com/597358816/Distilled-RL.

cs.LG↗

arXiv2026

SCOPE-RL: Stable and Quantitative Control of Policy Entropy in RL Post-Training

Reinforcement learning (RL) is a key paradigm for post-training large language models (LLMs), but the widely used Group Relative Policy Optimization (GRPO) often suffers from entropy collapse: exploration quickly disappears, policies converge prematurely, and sample diversity declines, ultimately harming training effectiveness. Existing remedies, including entropy bonuses and clip-based methods, rarely keep entropy within a stable exploration regime and often introduce oscillatory entropy or reward degradation. In this work, we identify a previously overlooked asymmetry in entropy dynamics: under high-temperature sampling, positive and negative samples have opposite effects on policy entropy. Specifically, high-temperature positive samples promote entropy growth, whereas negative samples suppress it. We provide a theoretical explanation for this phenomenon: when entropy decreases during policy updates, its derivative with respect to temperature is strictly positive under positive-sample updates, indicating that high-temperature positive samples can counteract entropy decay, thereby slowing entropy collapse and potentially reversing it. Motivated by this insight, we propose SCOPE-RL, a stable and quantitative entropy control framework through a regularization term constructed from temperature-adaptive positive samples. Extensive experiments show that SCOPE-RL consistently outperforms strong RL baselines on both Pass@1 and Pass@$k$. Our results provide evidence that escaping entropy collapse can improve reasoning performance, while also showing that the benefit is non-monotonic, with an optimal level of exploration for RL post-training in reasoning LLMs.

cs.LG↗

arXiv2026

Implicit Compression Regularization: Concise Reasoning via Internal Shorter Distributions in RL Post-Training

Reinforcement learning with verifiable rewards improves LLM reasoning but often induces overthinking, where models generate unnecessarily long reasoning traces. Existing methods mainly rely on length penalties or early-exit strategies; however, the former may degrade accuracy and induce underthinking, whereas the latter assumes that substantial portions of reasoning traces can be safely truncated. To obtain a compression signal without these limitations, we revisit the training dynamics of existing compression methods. We observe that the length--accuracy correlation is initially negative but continually increases during compression, indicating that shorter responses are initially more likely to be correct but gradually lose this property as the policy moves toward underthinking. Based on this observation, we formalize overthinking: a negative correlation indicates an overthinking regime, while a positive one indicates underthinking. When overthinking, the shortest correct responses are shorter than the group-average response length in expectation, making them natural compression targets already present in on-policy rollouts. We therefore propose \emph{Implicit Compression Regularization} (ICR), an on-policy regularization method whose compression signal comes from a virtual shorter distribution induced by the shortest correct responses in rollout groups, guiding the policy toward concise yet correct trajectories. Training dynamics show that ICR maintains a better length--accuracy correlation during compression, indicating that short responses remain better aligned with correctness instead of drifting toward underthinking. Experiments on three reasoning backbones and multiple mathematical and knowledge-intensive benchmarks show that ICR consistently shortens responses while preserving or improving accuracy, achieving a stronger accuracy--length Pareto frontier.

cs.AI↗

arXiv2026

Towards a Theoretical Understanding to the Generalization of RLHF

Reinforcement Learning from Human Feedback (RLHF) and its variants have emerged as the dominant approaches for aligning Large Language Models with human intent. While empirically effective, the theoretical generalization properties of these methods in high-dimensional settings remain to be explored. To this end, we build the generalization theory on RLHF of LLMs under the linear reward model, through the framework of algorithmic stability. In contrast to the existing works built upon the consistency of maximum likelihood estimations on reward model, our analysis is presented under an end-to-end learning framework, which is consistent with practice. Concretely, we prove that under a key \textbf{feature coverage} condition, the empirical optima of policy model have a generalization bound of order $\mathcal{O}(n^{-\frac{1}{2}})$. Moreover, the results can be extrapolated to parameters obtained by gradient-based learning algorithms, i.e., Gradient Ascent (GA) and Stochastic Gradient Ascent (SGA). Thus, we argue that our results provide new theoretical evidence for the empirically observed generalization of LLMs after RLHF.

cs.LG↗

arXiv2026

Distribution-Centric Policy Optimization Dominates Exploration-Exploitation Trade-off

The exploration-exploitation (EE) trade-off is a central challenge in reinforcement learning (RL) for large language models (LLMs). With Group Relative Policy Optimization (GRPO), training tends to be exploitation driven: entropy decreases monotonically, samples convergence, and exploration fades. Most existing fixes are \textbf{sample-centric}: they seek or bonus rare samples, assuming exploration comes from novel trajectories and tokens. These heuristics depend on the "luck" of informative samples, lack principled control of the policy, and often yield limited or inconsistent gains. In this work, we are the first to introduce a \textbf{distribution-centric} perspective for RL, in which exploration is always guided by a "better" target distribution, and reveal that a policy's ability to resist entropy collapse is governed by the distribution itself rather than individual samples. Building on this insight, we propose Distribution-Centric Policy Optimization (DCPO), which reformulates entropy regulation as distribution-level regularization. DCPO achieves controllable entropy fully on-policy without sampling from external distributions, enabling efficient exploration while maintaining training stability. Across multiple models and seven benchmarks, DCPO improves over GRPO by about 20\% on average. Overall, DCPO replaces sample-level heuristics with distribution-level principles, offering a theoretically grounded and flexible framework for controllable exploration and a stronger EE trade-off. The code is available in https://github.com/597358816/DCPO.

cs.LG↗