arXiv subjects

Chao Wang

Publications and source records attributed to Chao Wang.

At least 217 records · Page 12Linked to original sources

arXiv2025

TokenSelect: Efficient Long-Context Inference and Length Extrapolation for LLMs via Dynamic Token-Level KV Cache Selection

Rapid advances in Large Language Models (LLMs) have spurred demand for processing extended context sequences in contemporary applications. However, this progress faces two challenges: performance degradation due to sequence lengths out-of-distribution, and excessively long inference times caused by the quadratic computational complexity of attention. These issues limit LLMs in long-context scenarios. In this paper, we propose Dynamic Token-Level KV Cache Selection (TokenSelect), a training-free method for efficient and accurate long-context inference. TokenSelect builds upon the observation of non-contiguous attention sparsity, using QK dot products to measure per-head KV Cache criticality at token-level. By per-head soft voting mechanism, TokenSelect selectively involves a few critical KV cache tokens in attention calculation without sacrificing accuracy. To further accelerate TokenSelect, we design the Selection Cache based on observations of consecutive Query similarity and implemented the efficient Paged Dot Product Kernel, significantly reducing the selection overhead. A comprehensive evaluation of TokenSelect demonstrates up to $23.84\times$ speedup in attention computation and up to $2.28\times$ acceleration in end-to-end latency, while providing superior performance compared to state-of-the-art long-context inference methods.

cs.CL↗

arXiv2025

VideoMiner: Iteratively Grounding Key Frames of Hour-Long Videos via Tree-based Group Relative Policy Optimization

Understanding hour-long videos with multi-modal large language models (MM-LLMs) enriches the landscape of human-centered AI applications. However, for end-to-end video understanding with LLMs, uniformly sampling video frames results in LLMs being overwhelmed by a vast amount of irrelevant information as video length increases. Existing hierarchical key frame extraction methods improve the accuracy of video understanding but still face two critical challenges. 1) How can the interference of extensive redundant information in long videos be mitigated? 2) How can a model dynamically adapt to complex hierarchical structures while accurately identifying key frames? To address these issues, we propose VideoMiner, which iteratively segments, captions, and clusters long videos, forming a hierarchical tree structure. The proposed VideoMiner progresses from long videos to events to frames while preserving temporal coherence, effectively addressing the first challenge. To precisely locate key frames, we introduce T-GRPO, a tree-based group relative policy optimization in reinforcement learning method that guides the exploration of the VideoMiner. The proposed T-GRPO is specifically designed for tree structures, integrating spatiotemporal information at the event level while being guided by the question, thus solving the second challenge. We achieve superior performance in all long-video understanding tasks and uncover several interesting insights. Our proposed T-GRPO surprisingly incentivizes the model to spontaneously generate a reasoning chain. Additionally, the designed tree growth auxin dynamically adjusts the expansion depth, obtaining accuracy and efficiency gains. The code is publicly available at https://github.com/caoxinye/VideoMiner.

cs.CV↗

arXiv2025

Integrated photonic platform with high-speed entanglement generation and witnessing

High-speed generation and efficient entanglement detection on a photonic chip are essential for quantum information applications but hard to achieve due to common photonic chips' material properties and limited component performance. In this work, we experimentally demonstrate entanglement witness on a silicon photonic chip, with multi-rail single-photon entanglement generation based on decoy-state techniques. The detection is based on balanced homodyne detectors on the same photonic chip with a bandwidth of up to 12.5 GHz, which allows room-temperature operation. A loss-equivalent analysis method compensates for optical losses and system noises. Experimental results quantify an entangled state fidelity of 92% in quantum state tomography and a Clauser-Horne-Shimony-Holt (CHSH) violation lower bound of 2.59. These results establish a viable path toward fully integrated, high-bandwidth, room-temperature quantum photonic systems, with potential applications in on-chip quantum optics and quantum random number generation.

quant-ph↗

arXiv2025

Understanding Formal Reasoning Failures in LLMs as Abstract Interpreters

Large language models (LLMs) are increasingly used for program verification, and yet little is known about \emph{how} they reason about program semantics during this process. In this work, we focus on abstract interpretation based-reasoning for invariant generation and introduce two novel prompting strategies that aim to elicit such reasoning from LLMs. We evaluate these strategies across several state-of-the-art LLMs on 22 programs from the SV-COMP benchmark suite widely used in software verification. We analyze both the soundness of the generated invariants and the key thematic patterns in the models' reasoning errors. This work aims to highlight new research opportunities at the intersection of LLMs and program verification for applying LLMs to verification tasks and advancing their reasoning capabilities in this application.

cs.LG↗

arXiv2025

From Faithfulness to Correctness: Generative Reward Models that Think Critically

Through reinforcement learning with verifiable rewards (RLVR), large language models have achieved substantial progress in domains with easily verifiable outcomes, such as mathematics and coding. However, when applied to more complex tasks like open-domain question answering, RLVR faces significant challenges due to the difficulty of verifying correctness. The nuanced and ambiguous nature of real-world knowledge makes it difficult to reliably evaluate correctness in these settings, necessitating further abilities that extend beyond mere logical consistency to encompass an understanding and assessment of both external and internal knowledge. Recent work has primarily focused on improving faithfulness, defined as semantic alignment with supporting documents, which can cause models to rely excessively on external sources and diminish their capacity for critical assessment. To address this, we propose the Thinking-supervised Reward Model (TRM), which incorporates sentence-level thinking supervision to endow reward models with critical thinking abilities. Given a query, answer, and supporting documents, TRM first assesses the faithfulness of each answer sentence to the supporting documents, and then applies a reasoning step to evaluate sentence-level correctness. By structuring reward modeling as a sequence of faithfulness, reasoning, and correctness evaluations, TRM encourages models to critically assess and leverage both external and internal knowledge. Experiments on reward signals demonstrate that TRM substantially improves the identification of incorrect sentences, and incorporating TRM into policy optimization leads to significant gains in both answer correctness and usefulness.

cs.CL↗

arXiv2025

Understanding Textual Capability Degradation in Speech LLMs via Parameter Importance Analysis

The integration of speech into Large Language Models (LLMs) has substantially expanded their capabilities, but often at the cost of weakening their core textual competence. This degradation limits the ability of speech-enabled LLMs to fully exploit their pre-trained text-based knowledge. In this work, we analyze the underlying mechanisms of this issue through a focused study of the widely used encoder-adaptor paradigm. We propose an analytical framework based on parameter importance estimation, which reveals that fine-tuning for speech introduces a textual importance distribution shift: the layer-wise allocation of parameters critical to textual reasoning is disrupted. Building on this insight, we investigate two mitigation strategies: layer-wise learning rate scheduling and Low-Rank Adaptation (LoRA), both aim to preserve the original parameter distribution. Experimental results show that both approaches better maintain textual competence than full fine-tuning, while also improving downstream spoken question answering performance. Furthermore, our analysis offers a principled explanation for the effectiveness of the proposed mitigation strategies, linking their benefits to the structural properties of textual knowledge in LLMs.

cs.CL↗

arXiv2025

Learning More with Less: A Dynamic Dual-Level Down-Sampling Framework for Efficient Policy Optimization

Critic-free methods like GRPO reduce memory demands by estimating advantages from multiple rollouts but tend to converge slowly, as critical learning signals are diluted by an abundance of uninformative samples and tokens. To tackle this challenge, we propose the \textbf{Dynamic Dual-Level Down-Sampling (D$^3$S)} framework that prioritizes the most informative samples and tokens across groups to improve the efficient of policy optimization. D$^3$S operates along two levels: (1) the sample-level, which selects a subset of rollouts to maximize advantage variance ($\text{Var}(A)$). We theoretically proven that this selection is positively correlated with the upper bound of the policy gradient norms, yielding higher policy gradients. (2) the token-level, which prioritizes tokens with a high product of advantage magnitude and policy entropy ($|A_{i,t}|\times H_{i,t}$), focusing updates on tokens where the policy is both uncertain and impactful. Moreover, to prevent overfitting to high-signal data, D$^3$S employs a dynamic down-sampling schedule inspired by curriculum learning. This schedule starts with aggressive down-sampling to accelerate early learning and gradually relaxes to promote robust generalization. Extensive experiments on Qwen2.5 and Llama3.1 demonstrate that integrating D$^3$S into advanced RL algorithms achieves state-of-the-art performance and generalization while requiring \textit{fewer} samples and tokens across diverse reasoning benchmarks. Our code is added in the supplementary materials and will be made publicly available.

cs.LG↗

arXiv2025

DAMR: Efficient and Adaptive Context-Aware Knowledge Graph Question Answering with LLM-Guided MCTS

Knowledge Graph Question Answering (KGQA) aims to interpret natural language queries and perform structured reasoning over knowledge graphs by leveraging their relational and semantic structures to retrieve accurate answers. Existing methods primarily follow either the retrieve-then-reason paradigm, which relies on Graph Neural Networks or heuristic rules to extract static candidate paths, or dynamic path generation strategies that employ LLMs with prompting to jointly perform retrieval and reasoning. However, the former lacks adaptability due to static path extraction and the absence of contextual refinement, while the latter suffers from high computational costs and limited evaluation accuracy because of their dependence on fixed scoring functions and repeated LLM calls. To address these issues, this paper proposes Dynamically Adaptive MCTS-based Reasoning (DAMR), a novel framework that integrates LLM-guided Monte Carlo Tree Search (MCTS) with adaptive path evaluation to enable efficient and context-aware KGQA. DAMR leverages MCTS as a backbone, where an LLM-based planner selects the top-$k$ semantically relevant relations at each expansion step to effectively reduce the search space. To enhance evaluation accuracy, we introduce a lightweight Transformer-based scorer that performs context-aware plausibility estimation by jointly encoding the question and relation sequence through cross-attention, thereby capturing fine-grained semantic shifts during multi-hop reasoning. Furthermore, to mitigate the scarcity of high-quality supervision, DAMR incorporates a dynamic pseudo-path refinement mechanism that periodically generates training signals from partial paths explored during search, enabling the scorer to continually adapt to the evolving distribution of reasoning trajectories. Extensive experiments on multiple KGQA benchmarks show that DAMR significantly outperforms SOTA methods.

cs.CL↗

arXiv2025

QCD factorization for the $B\to γ\ellν_{\ell}$ decay beyond leading power

The radiative leptonic $B\to γ\ellν_{\ell}$ decay serves as an ideal platform to determine the $B$-meson inverse moment which is a fundamental nonperturbative parameter for the $B$ meson. In this paper, we explore precise QCD contributions to this decay with an energetic photon. We reproduce the next-to-next-to-leading-logarithmic resummation formula for the decay amplitude at leading power in $Λ_{\rm QCD}/m_b$. Employing operator identities, we calculate subleading-power contributions from the expansion of the hard-collinear propagator of the internal up quark and the heavy-quark expansion of the bottom quark. We update the contributions from the hadronic structure of the photon to the $\decay$ process with the dispersion technique. Together with other yet known power corrections, phenomenological applications including the partial branching fraction and ratio of the branching fractions of the radiative $B$ decay are investigated.

hep-ph↗

arXiv2025

Quantum Computational Insurance and Actuarial Science

In recent years, quantum computation has been rapidly advancing, driving a technological revolution with significant potential across various sectors, particularly in finance. Despite this, the insurance industry, an essential tool for mitigating unforeseen risks and losses, has received limited attention. This paper provides an initial exploration into the realm of quantum computational insurance and actuarial science. After introducing key insurance models and challenges, we discuss quantum algorithms that can address insurance problems based on their mathematical nature. Our study includes experimental and numerical demonstrations of quantum applications in non-life insurance, life insurance, and reinsurance. Additionally, we explore the timeline for quantum insurance, the development of quantum-enhanced insurance products, and the challenges posed by quantum computational advancements. This work systematically constructs the connection between quantum computation and the insurance industry, enhancing the development of insurance while promoting the application of quantum computation to more realistic problems.

quant-ph↗

arXiv2025

Kernel Ridge Regression with Predicted Feature Inputs and Applications to Factor-Based Nonparametric Regression

Kernel methods, particularly kernel ridge regression (KRR), are time-proven, powerful nonparametric regression techniques known for their rich capacity, analytical simplicity, and computational tractability. The analysis of their predictive performance has received continuous attention for more than two decades. However, in many modern regression problems where the feature inputs used in KRR cannot be directly observed and must instead be inferred from other measurements, the theoretical foundations of KRR remain largely unexplored. In this paper, we introduce a novel approach for analyzing KRR with predicted feature inputs. Our framework is not only essential for handling predicted feature inputs -- enabling us to derive risk bounds without imposing any assumptions on the error of the predicted feature -- but also strengthens existing analyses in the classical setting by allowing arbitrary model misspecification, requiring weaker conditions under the squared loss, particularly allowing both an unbounded response and an unbounded function class, and being flexible enough to accommodate other convex loss functions. We apply our general theory to factor-based nonparametric regression models and establish the minimax optimality of KRR when the feature inputs are predicted using principal component analysis. Our theoretical findings are further corroborated by simulation studies and real-data analyses using pretrained LLM embeddings for the downstream prediction task.

math.ST↗

arXiv2025

AIM 2025 challenge on Inverse Tone Mapping Report: Methods and Results

This paper presents a comprehensive review of the AIM 2025 Challenge on Inverse Tone Mapping (ITM). The challenge aimed to push forward the development of effective ITM algorithms for HDR image reconstruction from single LDR inputs, focusing on perceptual fidelity and numerical consistency. A total of \textbf{67} participants submitted \textbf{319} valid results, from which the best five teams were selected for detailed analysis. This report consolidates their methodologies and performance, with the lowest PU21-PSNR among the top entries reaching 29.22 dB. The analysis highlights innovative strategies for enhancing HDR reconstruction quality and establishes strong benchmarks to guide future research in inverse tone mapping.

cs.CV↗

arXiv2025

When Color-Space Decoupling Meets Diffusion for Adverse-Weather Image Restoration

Adverse Weather Image Restoration (AWIR) is a highly challenging task due to the unpredictable and dynamic nature of weather-related degradations. Traditional task-specific methods often fail to generalize to unseen or complex degradation types, while recent prompt-learning approaches depend heavily on the degradation estimation capabilities of vision-language models, resulting in inconsistent restorations. In this paper, we propose \textbf{LCDiff}, a novel framework comprising two key components: \textit{Lumina-Chroma Decomposition Network} (LCDN) and \textit{Lumina-Guided Diffusion Model} (LGDM). LCDN processes degraded images in the YCbCr color space, separately handling degradation-related luminance and degradation-invariant chrominance components. This decomposition effectively mitigates weather-induced degradation while preserving color fidelity. To further enhance restoration quality, LGDM leverages degradation-related luminance information as a guiding condition, eliminating the need for explicit degradation prompts. Additionally, LGDM incorporates a \textit{Dynamic Time Step Loss} to optimize the denoising network, ensuring a balanced recovery of both low- and high-frequency features in the image. Finally, we present DriveWeather, a comprehensive all-weather driving dataset designed to enable robust evaluation. Extensive experiments demonstrate that our approach surpasses state-of-the-art methods, setting a new benchmark in AWIR. The dataset and code are available at: https://github.com/fiwy0527/LCDiff.

cs.CV↗

arXiv2025

LongCat-Flash Technical Report

We introduce LongCat-Flash, a 560-billion-parameter Mixture-of-Experts (MoE) language model designed for both computational efficiency and advanced agentic capabilities. Stemming from the need for scalable efficiency, LongCat-Flash adopts two novel designs: (a) Zero-computation Experts, which enables dynamic computational budget allocation and activates 18.6B-31.3B (27B on average) per token depending on contextual demands, optimizing resource usage. (b) Shortcut-connected MoE, which enlarges the computation-communication overlap window, demonstrating notable gains in inference efficiency and throughput compared to models of a comparable scale. We develop a comprehensive scaling framework for large models that combines hyperparameter transfer, model-growth initialization, a multi-pronged stability suite, and deterministic computation to achieve stable and reproducible training. Notably, leveraging the synergy among scalable architectural design and infrastructure efforts, we complete model training on more than 20 trillion tokens within 30 days, while achieving over 100 tokens per second (TPS) for inference at a cost of \$0.70 per million output tokens. To cultivate LongCat-Flash towards agentic intelligence, we conduct a large-scale pre-training on optimized mixtures, followed by targeted mid- and post-training on reasoning, code, and instructions, with further augmentation from synthetic data and tool use tasks. Comprehensive evaluations demonstrate that, as a non-thinking foundation model, LongCat-Flash delivers highly competitive performance among other leading models, with exceptional strengths in agentic tasks. The model checkpoint of LongCat-Flash is open-sourced to foster community research. LongCat Chat: https://longcat.ai Hugging Face: https://huggingface.co/meituan-longcat GitHub: https://github.com/meituan-longcat

cs.CL↗

arXiv2025

CIDER: A Causal Cure for Brand-Obsessed Text-to-Image Models

Text-to-image (T2I) models exhibit a significant yet under-explored "brand bias", a tendency to generate contents featuring dominant commercial brands from generic prompts, posing ethical and legal risks. We propose CIDER, a novel, model-agnostic framework to mitigate bias at inference-time through prompt refinement to avoid costly retraining. CIDER uses a lightweight detector to identify branded content and a Vision-Language Model (VLM) to generate stylistically divergent alternatives. We introduce the Brand Neutrality Score (BNS) to quantify this issue and perform extensive experiments on leading T2I models. Results show CIDER significantly reduces both explicit and implicit biases while maintaining image quality and aesthetic appeal. Our work offers a practical solution for more original and equitable content, contributing to the development of trustworthy generative AI.

cs.CV↗

arXiv2025

Graph Signal Processing for Global Stock Market Realized Volatility Forecasting

This paper introduces an innovative realized volatility (RV) forecasting framework that extends the conventional Heterogeneous autoregressive (HAR) model via integrating Graph Signal Processing (GSP). The study first evaluates various constructions of volatility-interrelationship networks by analyzing how the associated graph signal energy tracks global financial market volatility. Volatility spillovers are subsequently embedded into the proposed framework, which employs the graph Fourier transform (GFT) and its inverse to effectively capture global stock market dynamics in both the spectral and spatial domains. The framework not only provides a global context for modeling the volatility interrelationships, but also captures the nonlinearity and directionality of the volatility spillover effect. The empirical study using RV data of $24$ global stock market indices compares short-, mid- and long-term RV forecasts with various HAR-type benchmarks and a graph neural network-based HAR model. The proposed model consistently outperforms all comparators, demonstrating the effectiveness of integrating GSP into the HAR model for RV forecasting.

q-fin.GN↗

arXiv2025

Evaluating Supervised Learning Models for Fraud Detection: A Comparative Study of Classical and Deep Architectures on Imbalanced Transaction Data

Fraud detection remains a critical task in high-stakes domains such as finance and e-commerce, where undetected fraudulent transactions can lead to significant economic losses. In this study, we systematically compare the performance of four supervised learning models - Logistic Regression, Random Forest, Light Gradient Boosting Machine (LightGBM), and a Gated Recurrent Unit (GRU) network - on a large-scale, highly imbalanced online transaction dataset. While ensemble methods such as Random Forest and LightGBM demonstrated superior performance in both overall and class-specific metrics, Logistic Regression offered a reliable and interpretable baseline. The GRU model showed strong recall for the minority fraud class, though at the cost of precision, highlighting a trade-off relevant for real-world deployment. Our evaluation emphasizes not only weighted averages but also per-class precision, recall, and F1-scores, providing a nuanced view of each model's effectiveness in detecting rare but consequential fraudulent activity. The findings underscore the importance of choosing models based on the specific risk tolerance and operational needs of fraud detection systems.

cs.LG↗

arXiv2025

Global Stock Market Volatility Forecasting Incorporating Dynamic Graphs and All Trading Days

This paper introduces a global stock market volatility forecasting model that enhances forecasting accuracy and practical utility in real-world financial decision-making by integrating dynamic graph structures and encompassing all active trading days of different stock markets. The model employs a spatial-temporal graph neural network architecture to capture the volatility spillover effect, where shocks in one market spread to others through the interconnective global economy. By calculating the volatility spillover index to depict the volatility network as graphs, the model effectively mirrors the volatility dynamics for the chosen stock market indices. In the empirical analysis covering 8 global market indices, the realized volatility forecasting performance of the proposed model surpasses the baseline models in all forecasting scenarios.

q-fin.GN↗