Zixuan Zhong

Zhongwei Wan, Yun Shen, Zhihao Dou et al.

Reinforcement learning with verifiers (RLVR) is a central paradigm for improving large language model (LLM) reasoning, yet existing methods often suffer from limited exploration. Policies tend to collapse onto a few reasoning patterns and prematurely stop deep exploration, while conventional entropy regularization introduces only local stochasticity and fails to induce meaningful path-level diversity, leading to weak and unstable learning signals in group-based policy optimization. We propose DSDR, a Dual-Scale Diversity Regularization reinforcement learning framework that decomposes diversity in LLM reasoning into global and coupling components. Globally, DSDR promotes diversity among correct reasoning trajectories to explore distinct solution modes. Locally, it applies a length-invariant, token-level entropy regularization restricted to correct trajectories, preventing entropy collapse within each mode while preserving correctness. The two scales are coupled through a global-to-local allocation mechanism that emphasizes local regularization for more distinctive correct trajectories. We provide theoretical support showing that DSDR preserves optimal correctness under bounded regularization, sustains informative learning signals in group-based optimization, and yields a principled global-to-local coupling rule. Experiments on multiple reasoning benchmarks demonstrate consistent improvements in accuracy and pass@k, highlighting the importance of dual-scale diversity for deep exploration in RLVR. Code is available at https://github.com/SUSTechBruce/DSDR.

Tongshun Zhang, Pingping Liu, Yuqing Lei et al.

Limited illumination often causes severe physical noise and detail degradation in images. Existing Low-Light Image Enhancement (LLIE) methods frequently treat the enhancement process as a blind black-box mapping, overlooking the physical noise transformation during imaging, leading to suboptimal performance. To address this, we propose a novel LLIE approach, conceptually formulated as a physics-based attack and display-adaptive defense paradigm. Specifically, on the attack side, we establish a physics-based Degradation Synthesis (PDS) pipeline. Unlike standard data augmentation, PDS explicitly models Image Signal Processor (ISP) inversion to the RAW domain, injects physically plausible photon and read noise, and re-projects the data to the sRGB domain. This generates high-fidelity training pairs with explicitly parameterized degradation vectors, effectively simulating realistic attacks on clean signals. On the defense side, we construct a dual-layer fortified system. A noise predictor estimates degradation parameters from the input sRGB image. These estimates guide a degradation-aware Mixture of Experts (DA-MoE), which dynamically routes features to experts specialized in handling specific noise intensities. Furthermore, we introduce an Adaptive Metric Defense (AMD) mechanism, dynamically calibrating the feature embedding space based on noise severity, ensuring robust representation learning under severe degradation. Extensive experiments demonstrate that our approach offers significant plug-and-play performance enhancement for existing benchmark LLIE methods, effectively suppressing real-world noise while preserving structural fidelity. The sourced code is available at https://github.com/bywlzts/Attack-defense-llie.

Yingtie Lei, Zhongwei Wan, Jiankun Zhang et al.

Large language model (LLM) agents accumulate rich episodic trajectories while solving real-world tasks, but it remains unclear whether such experience can be distilled into reusable procedural skills. We introduce SkillEvolBench, a diagnostic benchmark for evaluating this step from experience reuse to skill formation. It contains 180 tasks across six real-world agent environments, organized into role-conditioned task families with shared latent procedures. Agents learn from acquisition tasks, update an external skill library using compacted trajectories and verifier feedback, and then face frozen deployment tasks testing context shift, adversarial shortcuts, and composition. By comparing self-generated and curated-start skill evolution against no-skill and raw-trajectory controls, SkillEvolBench separates procedural abstraction from base capability, curated prior knowledge, and direct reuse of episodic traces. Across ten model configurations and three agent harnesses, we find that current agents often adapt locally but rarely form robust reusable skills. Skill-based conditions can improve acquisition or replay, and individual models sometimes gain on specific deployment axes, but these gains are unstable under frozen deployment. Raw-trajectory reuse frequently outperforms distilled skills, suggesting that current abstraction procedures discard contextual and procedural cues that remain useful for future tasks. Capacity and cost analyses further show that writing more skills or larger Tier-3 resource libraries is not sufficient: additional updates can improve coverage while introducing episode-specific drift and procedural clutter. These findings position SkillEvolBench as a testbed for measuring when one-off experience becomes durable procedural knowledge rather than task-local memory.

Tongshun Zhang, Pingping Liu, Zixuan Zhong et al.

Recovering fine-grained details in extremely dark images remains challenging due to severe structural information loss and noise corruption. Existing enhancement methods often fail to preserve intricate details and sharp edges, limiting their effectiveness in downstream applications like text and edge detection. To address these deficiencies, we propose an efficient dual-stage approach centered on detail recovery for dark images. In the first stage, we introduce a Residual Fourier-Guided Module (RFGM) that effectively restores global illumination in the frequency domain. RFGM captures inter-stage and inter-channel dependencies through residual connections, providing robust priors for high-fidelity frequency processing while mitigating error accumulation risks from unreliable priors. The second stage employs complementary Mamba modules specifically designed for textural structure refinement: (1) Patch Mamba operates on channel-concatenated non-downsampled patches, meticulously modeling pixel-level correlations to enhance fine-grained details without resolution loss. (2) Grad Mamba explicitly focuses on high-gradient regions, alleviating state decay in state space models and prioritizing reconstruction of sharp edges and boundaries. Extensive experiments on multiple benchmark datasets and downstream applications demonstrate that our method significantly improves detail recovery performance while maintaining efficiency. Crucially, the proposed modules are lightweight and can be seamlessly integrated into existing Fourier-based frameworks with minimal computational overhead. Code is available at https://github.com/bywlzts/RFGM.