Reinforcement Learning with V erifiable Reward (RL VR) has significantly advanced the complex reasoning abilities of Large Language Models (LLMs). However, it struggles to break through the inherent capability boundaries of the base LLM, due to its essentially on-policy strategy coupled with LLM's immense action space and sparse reward. Critically, RL VR can lead to the capability boundary collapse, narrowing the LLM's problem-solving scope. To address this problem, we propose RL-PLUS, a novel hybrid-policy optimization approach for LLMs that synergizes internal exploitation with external data to achieve stronger reasoning capabilities and surpass the boundaries of base models. RL-PLUS integrates two core components, i.e., Multiple Importance Sampling to address distributional mismatch from external data, and Exploration-Based Advantage Function to guide the model towards high-value, unexplored reasoning paths. We provide both theoretical analysis and extensive experiments to demonstrate the superiority and gen-eralizability of our approach. Compared with existing RL VR methods, RL-PLUS achieves 1) state-of-the-art performance on six math reasoning benchmarks; 2) superior performance on six out-of-distribution reasoning tasks; 3) consistent and significant gains across diverse model families, with average relative improvements up to 69.2%. Moreover, the analysis of Pass@k curves indicates that RL-PLUS effectively resolves the capability boundary collapse problem.

Abstract--Background: Automated analysis of lung sound auscultation is essential for monitoring respiratory health, particularly in regions with a shortage of skilled healthcare workers. Although respiratory sound classification has been widely studied in adults, its application in pediatric populations, especially in children under six years of age remains underexplored. Developmental changes in pediatric lungs substantially modify the acoustic properties of respiratory sounds, requiring classification approaches tailored specifically to this age group. Methods: T o address this challenge, we propose a multistage hybrid CNN-Transformer framework that integrates CNN-extracted features with an attention-based architecture for pediatric respiratory disease classification. Scalogram images were generated from both full recordings and individual breath events to capture multi-resolution representations of respiratory sounds. T o mitigate class imbalance, class-wise focal loss was applied during model training. Results: The proposed model achieved an overall score of 0.9039 in binary event classification At the recording level, the model obtained scores of 0.720 for ternary classification and 0.571 for multiclass classification. These results outperform the previous best-performing models by 3.81% and 5.94%, respectively. Conclusion: Our findings demonstrate that the proposed hybrid CNN-Transformer framework effectively captures the unique acoustic features of pediatric lung sounds.

This study positions large language models (LLMs) as "subjective literary critics" to explore aesthetic preferences and evaluation patterns in literary assessment. Ten Japanese science fiction short stories were translated into English and evaluated by six state-of-the-art LLMs across seven independent sessions. Principal component analysis and clustering techniques revealed significant variations in evaluation consistency (α ranging from 1.00 to 0.35) and five distinct evaluation patterns. Additionally, evaluation variance across stories differed by up to 4.5-fold, with TF-IDF analysis confirming distinctive evaluation vocabularies for each model. Our seven-session within-day protocol using an original Science Fiction corpus strategically minimizes external biases, allowing us to observe implicit value systems shaped by RLHF and their influence on literary judgment. These findings suggest that LLMs may possess individual evaluation characteristics similar to human critical schools, rather than functioning as neutral benchmarkers.

Recent studies suggest that large language models (LLMs) possess the capability to solve graph reasoning tasks. Notably, even when graph structures are embedded within textual descriptions, LLMs can still effectively answer related questions. This raises a fundamental question: How can a decoder-only Transformer architecture understand underlying graph structures? To address this, we start with the substructure extraction task, interpreting the inner mechanisms inside the transformers and analyzing the impact of the input queries. Specifically, through both empirical results and theoretical analysis, we present Induced Substructure Filtration (ISF), a perspective that captures the substructure identification in the multi-layer transformers. We further validate the ISF process in LLMs, revealing consistent internal dynamics across layers. Building on these insights, we explore the broader capabilities of Transformers in handling diverse graph types. Specifically, we introduce the concept of thinking in substructures to efficiently extract complex composite patterns, and demonstrate that decoder-only Transformers can successfully extract substructures from attributed graphs, such as molecular graphs. Together, our findings offer a new insight on how sequence-based Transformers perform the substructure extraction task over graph data.

Accurate brain tumor segmentation is crucial for neuro-oncology diagnosis and treatment planning. Deep learning methods have made significant progress, but automatic segmentation still faces challenges. These include tumor morphological heterogeneity and complex three-dimensional spatial relationships. This paper proposes a three-tier fusion architecture that achieves precise brain tumor segmentation. The method processes information progressively at the pixel, feature, and semantic levels. At the pixel level, physical modeling extends magnetic resonance imaging (MRI) to multimodal data. This includes simulated ultrasound and synthetic computed tomography (CT). At the feature level, the method achieves Transformer-based cross-modal feature fusion through multi-teacher collaborative distillation. At the semantic level, clinical textual knowledge generated by GPT-4V transforms into spatial guidance signals. This transformation uses Contrastive Language-Image Pre-training (CLIP) contrastive learning and Feature-wise Linear Modulation (FiLM). These three tiers work together to form a complete processing chain. The chain spans from data augmentation to feature extraction to semantic guidance. The model achieves average Dice coefficients of 0.8665, 0.9014, and 0.8912 on the three datasets, respectively. The 95% Hausdorff Distance (HD95) reduces by an average of 6.57 millimeters compared to the baseline. Introduction Magnetic resonance imaging (MRI) segmentation of brain tumors is fundamental for neuro-oncology diagnosis and treatment planning. Different sub-regions have different prognostic significance.

We address the task of zero-shot video classification for extremely fine-grained actions (e.g., Windmill Dunk in basketball), where no video examples or temporal annotations are available for unseen classes. While image-language models (e.g., CLIP, SigLIP) show strong open-set recognition, they lack temporal modeling needed for video understanding. We propose ActAlign, a truly zero-shot, training-free method that formulates video classification as a sequence alignment problem, preserving the generalization strength of pretrained image-language models. For each class, a large language model (LLM) generates an ordered sequence of sub-actions, which we align with video frames using Dynamic Time Warping (DTW) in a shared embedding space. Without any video-text supervision or fine-tuning, ActAlign achieves 30.5% accuracy on ActionAtlas--the most diverse benchmark of fine-grained actions across multiple sports--where human performance is only 61.6%. ActAlign outperforms billion-parameter video-language models while using 8x fewer parameters. Our approach is model-agnostic and domain-general, demonstrating that structured language priors combined with classical alignment methods can unlock the open-set recognition potential of image-language models for fine-grained video understanding.

Recent advancements in large language models (LLMs) have significantly advanced complex reasoning capabilities, particularly through extended chain-of-thought (CoT) reasoning that incorporates mechanisms such as backtracking, self-reflection, and self-correction. Despite these developments, the self-correction abilities of LLMs during long CoT reasoning remain underexplored. And recent findings on overthinking suggest that such models often engage in unnecessarily redundant reasoning. In this work, we empirically show that the first reasoning step exerts a disproportionately large influence on the final prediction. I.e., errors introduced at this stage can substantially degrade subsequent reasoning quality. This phenomenon is consistently observed across various state-of-the-art open- and closed-source reasoning models. Leveraging this insight, we propose an efficient sampling strategy that leverages a reward model to identify and retain high-quality first reasoning steps while discarding suboptimal ones, achieving up to a 70% reduction in inference cost without sacrificing any accuracy. Our work highlights the central role of the first reasoning step in generating a high-quality reasoning trajectory, and thus enabling significantly efficient sampling.

Quantization has emerged as an effective and lightweight solution to reduce the memory footprint of the KV cache in Large Language Models. Nevertheless, minimizing the accuracy degradation caused by ultra-low-bit KV cache quantization remains a significant challenge. While scalar quantization is constrained by 1-bit bound, vector quantization exploits intra-vector correlations and enables sub-bit regimes, making it more suitable for ultra-low-bit quantization. To further mitigate quantization-induced degradation, we reveal that the degradation is highly uneven across tokens in attention quality. To investigate this unevenness, we introduce anchor score to measure each token's sensitivity to quantization. Our analysis and experiments show that preserving a small subset (1\%) of tokens with the highest Anchor Score significantly mitigates accuracy loss under aggressive quantization. We propose AnTKV, a dual-stage framework that leverages anchor token-aware vector quantization to compress the KV cache. It combines offline token-aware centroids learning and online anchor token selection to balance compression and accuracy. To enable efficient deployment, we design an online anchor token selection kernel compatible with FlashAttention. It allows LLaMA3-8B to scale to 840K tokens on a single 80GB A100, while delivering up to $3.5\times$ higher decoding throughput over the FP16 baseline. Experiments demonstrate that AnTKV matches or surpasses prior methods at 4-bit, and significantly reduce perplexity under ultra-low-bit quantization, achieving 6.32 at 1-bit on Mistral-7B, compared to 7.25 for CQ and 15.36 for KVQuant.

The advent of artificial intelligence has propelled AI-Generated Game Commentary (AI-GGC) into a rapidly expanding field, offering benefits such as unlimited availability and personalized narration. However, current researches in this area remain fragmented, and a comprehensive survey that systematically unifies existing efforts is still missing. To bridge this gap, our survey introduces a unified framework that systematically organizes the AI-GGC landscape. We present a novel taxonomy focused on three core commentator capabilities: Live Observation, Strategic Analysis, and Historical Recall. Commentary is further categorized into three functional types: Descriptive, Analytical, and Background. Building on this structure, we provide an in-depth review of state-of-the-art methods, datasets, and evaluation metrics across various game genres. Finally, we highlight key challenges such as real-time reasoning, multimodal integration, and evaluation bottlenecks, and outline promising directions for future research and system development in AI-GGC.

Multimodal large language models (MLLMs) frequently hallucinate by over-committing to spurious visual cues. Prior remedies-Visual and Instruction Contrastive Decoding (VCD, ICD)-mitigate this issue, yet the mechanism remains opaque. We first empirically show that their improvements systematically coincide with redistributions of cross-modal attention. Building on this insight, we propose Attention-Steerable Contrastive Decoding (ASCD), which directly steers the attention scores during decoding. ASCD combines (i) positive steering, which amplifies automatically mined text-centric heads-stable within a model and robust across domains-with (ii) negative steering, which dampens on-the-fly identified critical visual tokens. The method incurs negligible runtime and memory overhead and requires no additional training. Across five MLLM backbones and three decoding schemes, ASCD reduces hallucination on POPE, CHAIR, and MMHal-Bench by up to 38.2 percent while improving accuracy on standard VQA benchmarks, including MMMU, MM-VET, ScienceQA, TextVQA, and GQA. These results position attention steering as a simple, model-agnostic, and principled route to safer, more faithful multimodal generation.