arXiv subjects

Zeyu Li

Publications and source records attributed to Zeyu Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

MINER: Mining Multimodal Internal Representation for Efficient Retrieval

Visual document retrieval has become essential for accessing information in visually rich documents. Existing approaches fall into two camps. Late-interaction retrievers achieve strong quality through fine-grained token-level matching but store hundreds of vectors per page, incurring large index footprints and high serving costs. By contrast, dense single-vector retrievers retain storage and latency advantages but consistently lag in quality because they compress all information into a single final-layer embedding. In this work, we first conduct a layerwise diagnostic on single-vector retrievers, revealing that retrieval-relevant signal resides in internal representations. Motivated by these findings, we propose MINER (Mining Multimodal Internal RepreseNtation for Efficient Retrieval), a lightweight plug-in module that probes and fuses internal signals across transformer layers into a single compact embedding without modifying the backbone or sacrificing single-vector efficiency. The first Retrieval-Aligned Layer Probing stage attaches a lightweight probe at each layer, surfacing which dimensions carry retrieval-relevant information. The subsequent Adaptive Sparse Multi-Layer Fusion stage applies performance-adaptive neuron-level masking to the selected layers and fuses the surviving signals into the final dense vector. Across ViDoRe V1/V2/V3, MINER outperforms existing dense single-vector retrievers on the majority of benchmarks, with up to 4.5% nDCG@5 improvement over its corresponding backbone. Compared to strong late-interaction baselines, in some settings MINER substantially narrows the nDCG@$5$ gap to $0.2$ while preserving the storage and serving advantages of dense retrieval.

cs.LG↗

arXiv2026

AcademiClaw: When Students Set Challenges for AI Agents

Benchmarks within the OpenClaw ecosystem have thus far evaluated exclusively assistant-level tasks, leaving the academic-level capabilities of OpenClaw largely unexamined. We introduce AcademiClaw, a bilingual benchmark of 80 complex, long-horizon tasks sourced directly from university students' real academic workflows -- homework, research projects, competitions, and personal projects -- that they found current AI agents unable to solve effectively. Curated from 230 student-submitted candidates through rigorous expert review, the final task set spans 25+ professional domains, ranging from olympiad-level mathematics and linguistics problems to GPU-intensive reinforcement learning and full-stack system debugging, with 16 tasks requiring CUDA GPU execution. Each task executes in an isolated Docker sandbox and is scored on task completion by multi-dimensional rubrics combining six complementary techniques, with an independent five-category safety audit providing additional behavioral analysis. Experiments on six frontier models show that even the best achieves only a 55\% pass rate. Further analysis uncovers sharp capability boundaries across task domains, divergent behavioral strategies among models, and a disconnect between token consumption and output quality, providing fine-grained diagnostic signals beyond what aggregate metrics reveal. We hope that AcademiClaw and its open-sourced data and code can serve as a useful resource for the OpenClaw community, driving progress toward agents that are more capable and versatile across the full breadth of real-world academic demands. All data and code are available at https://github.com/GAIR-NLP/AcademiClaw.

cs.AI↗

arXiv2026

Towards Hierarchical Multi-Step Reward Models for Enhanced Reasoning in Large Language Models

Recent studies show that Large Language Models (LLMs) achieve strong reasoning capabilities through supervised fine-tuning or reinforcement learning. However, a key approach, the Process Reward Model (PRM), suffers from reward hacking, making it unreliable in identifying the best intermediate step. In addition, the cost of annotating reasoning processes for reward modeling is high, making large-scale collection of high-quality data challenging. To address this, we propose a novel reward model approach called the Hierarchical Reward Model (HRM), which evaluates both individual and consecutive reasoning steps at both fine-grained and coarse-grained levels. HRM excels at assessing multi-step reasoning coherence, especially when flawed steps are later corrected through self-reflection. To further reduce the cost of generating training data, we introduce a lightweight and effective data augmentation strategy called Hierarchical Node Compression (HNC), which merges two consecutive reasoning steps into one within the tree structure. By applying HNC to MCTS-generated reasoning trajectories, we enhance the diversity and robustness of HRM training data while introducing controlled noise with minimal computational overhead. Empirical results on the PRM800K dataset show that HRM, together with HNC, provides more stable and reliable evaluations than PRM. Furthermore, cross-domain evaluations on the MATH500 and GSM8K datasets demonstrate HRM's strong generalization and robustness across a variety of reasoning tasks.

cs.CL↗

arXiv2026

WebArbiter: A Principle-Guided Reasoning Process Reward Model for Web Agents

Web agents hold great potential for automating complex computer tasks, yet their interactions involve long-horizon, sequential decision-making with irreversible actions. In such settings, outcome-based supervision is sparse and delayed, often rewarding incorrect trajectories and failing to support inference-time scaling. This motivates the use of Process Reward Models (WebPRMs) for web navigation, but existing approaches remain limited: scalar WebPRMs collapse progress into coarse, weakly grounded signals, while checklist-based WebPRMs rely on brittle template matching that fails under layout or semantic changes and often mislabels superficially correct actions as successful, providing little insight or interpretability. To address these challenges, we introduce WebArbiter, a reasoning-first, principle-inducing WebPRM that formulates reward modeling as text generation, producing structured justifications that conclude with a preference verdict and identify the action most conducive to task completion under the current context. Training follows a two-stage pipeline: reasoning distillation equips the model with coherent principle-guided reasoning, and reinforcement learning corrects teacher biases by directly aligning verdicts with correctness, enabling stronger generalization. To support systematic evaluation, we release WebPRMBench, a comprehensive benchmark spanning four diverse web environments with rich tasks and high-quality preference annotations. On WebPRMBench, WebArbiter-7B outperforms the strongest baseline, GPT-5, by 9.1 points. In reward-guided trajectory search on WebArena-Lite, it surpasses the best prior WebPRM by up to 6.4 points, underscoring its robustness and practical value in complex web tasks.

cs.AI↗

arXiv2026

NEURA: A Unified and Retargetable Compilation Framework for Coarse-Grained Reconfigurable Architectures

Coarse-Grained Reconfigurable Architectures (CGRAs) are a promising and versatile accelerator platform, offering a balance between the performance and efficiency of specialized accelerators and the software programmability. However, their full potential is severely hindered by control flow in accelerated kernels, as the control flow (e.g., loops, branches) is fundamentally incompatible with the parallel, data-driven CGRA fabric. Prior strategies to resolve this mismatch in CGRA kernel acceleration are either inefficient, sacrificing performance for generality, or lack generality due to the difficulty of adapting them across different execution models. Thus, a general and unified solution for efficient CGRA kernel acceleration remains elusive. This paper introduces NEURA, a unified and retargetable compilation framework that systematically resolves the control-dataflow mismatch in CGRAs. NEURA's core innovation is a novel, pure dataflow intermediate representation (IR) built on a predicated type system. In this IR, control contexts are embedded as a predicate within each data, making control an intrinsic property of data. This mechanism enables NEURA to systematically flatten complex control flow into a single unified dataflow graph. This unified representation decouples kernel representation from hardware, empowering NEURA to retarget diverse CGRAs with different execution models and microarchitectural features. When targeted to a high-performance spatio-temporal CGRA, NEURA delivers a 2.20x speedup on kernel benchmarks and up to 2.71x geometric mean speedup on real-world applications over state-of-the-art (SOTA) high-performance baselines. It also provides a competitive solution against the SOTA low-power CGRA when retargeted to a spatial-only CGRA. NEURA is open-source and available at https://github.com/coredac/neura.

cs.PL↗

arXiv2026

Tucker Diffusion Model for High-dimensional Tensor Generation

Statistical inference on large-dimensional tensor data has been extensively studied in the literature and widely used in economics, biology, machine learning, and other fields, but how to generate a structured tensor with a target distribution is still a new problem. As profound AI generators, diffusion models have achieved remarkable success in learning complex distributions. However, their extension to generating multi-linear tensor-valued observations remains underexplored. In this work, we propose a novel Tucker diffusion model for learning high-dimensional tensor distributions. We show that the score function admits a structured decomposition under the low Tucker rank assumption, allowing it to be both accurately approximated and efficiently estimated using a carefully tailored tensor-shaped architecture named Tucker-Unet. Furthermore, the distribution of generated tensors, induced by the estimated score function, converges to the true data distribution at a rate depending on the maximum of tensor mode dimensions, thereby offering a clear theoretical advantage over the naive vectorized approach, which has a product dependence. Empirically, compared to existing approaches, the Tucker diffusion model demonstrates strong practical potential in synthetic and real-world tensor generation tasks, achieving comparable and sometimes even superior statistical performance with significantly reduced training and sampling costs.

stat.ME↗

arXiv2026

PRBench: End-to-end Paper Reproduction in Physics Research

AI agents powered by large language models exhibit strong reasoning and problem-solving capabilities, enabling them to assist scientific research tasks such as formula derivation and code generation. However, whether these agents can reliably perform end-to-end reproduction from real scientific papers remains an open question. We introduce PRBench, a benchmark of 30 expert-curated tasks spanning 11 subfields of physics. Each task requires an agent to comprehend the methodology of a published paper, implement the corresponding algorithms from scratch, and produce quantitative results matching the original publication. Agents are provided only with the task instruction and paper content, and operate in a sandboxed execution environment. All tasks are contributed by domain experts from over 20 research groups at the School of Physics, Peking University, each grounded in a real published paper and validated through end-to-end reproduction with verified ground-truth results and detailed scoring rubrics. Using an agentified assessment pipeline, we evaluate a set of coding agents on PRBench and analyze their capabilities across key dimensions of scientific reasoning and execution. The best-performing agent, OpenAI Codex powered by GPT-5.3-Codex, achieves a mean overall score of 34%. All agents exhibit a zero end-to-end callback success rate, with particularly poor performance in data accuracy and code correctness. We further identify systematic failure modes, including errors in formula implementation, inability to debug numerical simulations, and fabrication of output data. Overall, PRBench provides a rigorous benchmark for evaluating progress toward autonomous scientific research.

cs.CL↗

arXiv2026

ZipServ: Fast and Memory-Efficient LLM Inference with Hardware-Aware Lossless Compression

Lossless model compression holds tremendous promise for alleviating the memory and bandwidth bottlenecks in bit-exact Large Language Model (LLM) serving. However, existing approaches often result in substantial inference slowdowns due to fundamental design mismatches with GPU architectures: at the kernel level, variable-length bitstreams produced by traditional entropy codecs break SIMT parallelism; at the system level, decoupled pipelines lead to redundant memory traffic. We present ZipServ, a lossless compression framework co-designed for efficient LLM inference. ZipServ introduces Tensor-Core-Aware Triple Bitmap Encoding (TCA-TBE), a novel fixed-length format that enables constant-time, parallel decoding, together with a fused decompression-GEMM (ZipGEMM) kernel that decompresses weights on-the-fly directly into Tensor Core registers. This "load-compressed, compute-decompressed" design eliminates intermediate buffers and maximizes compute intensity. Experiments show that ZipServ reduces the model size by up to 30%, achieves up to 2.21x kernel-level speedup over NVIDIA's cuBLAS, and expedites end-to-end inference by an average of 1.22x over vLLM. ZipServ is the first lossless compression system that provides both storage savings and substantial acceleration for LLM inference on GPUs.

cs.DC↗

arXiv2026

An End-to-end Architecture for Collider Physics and Beyond

We present, to our knowledge, the first language-driven agent system capable of executing end-to-end collider phenomenology tasks, instantiated within a decoupled, domain-agnostic architecture for autonomous High-Energy Physics phenomenology. Guided only by natural-language prompts supplemented with standard physics notation, ColliderAgent carries out workflows from a theoretical Lagrangian to final phenomenological outputs without relying on package-specific code. In this framework, a hierarchical multi-agent reasoning layer is coupled to Magnus, a unified execution backend for phenomenological calculations and simulation toolchains. We validate the system on representative literature reproductions spanning leptoquark and axion-like-particle scenarios, higher-dimensional effective operators, parton-level and detector-level analyses, and large-scale parameter scans leading to exclusion limits. These results point to a route toward more automated, scalable, and reproducible research in collider physics, cosmology, and physics more broadly.

hep-ph↗

arXiv2026

Constrained Particle Seeking: Solving Diffusion Inverse Problems with Just Forward Passes

Diffusion models have gained prominence as powerful generative tools for solving inverse problems due to their ability to model complex data distributions. However, existing methods typically rely on complete knowledge of the forward observation process to compute gradients for guided sampling, limiting their applicability in scenarios where such information is unavailable. In this work, we introduce \textbf{\emph{Constrained Particle Seeking (CPS)}}, a novel gradient-free approach that leverages all candidate particle information to actively search for the optimal particle while incorporating constraints aligned with high-density regions of the unconditional prior. Unlike previous methods that passively select promising candidates, CPS reformulates the inverse problem as a constrained optimization task, enabling more flexible and efficient particle seeking. We demonstrate that CPS can effectively solve both image and scientific inverse problems, achieving results comparable to gradient-based methods while significantly outperforming gradient-free alternatives. Code is available at https://github.com/deng-ai-lab/CPS.

cs.LG↗

arXiv2025

Aligned Anchor Groups Guided Line Segment Detector

This paper introduces a novel line segment detector, the Aligned Anchor Groups guided Line Segment Detector (AAGLSD), designed to detect line segments from images with high precision and completeness. The algorithm employs a hierarchical approach to extract candidate pixels with different saliency levels, including regular anchors and aligned anchor groups. AAGLSD initiates from these aligned anchor groups, sequentially linking anchors and updating the currently predicted line segment simultaneously. The final predictions are derived through straightforward validation and merging of adjacent line segments, avoiding complex refinement strategies. AAGLSD is evaluated on various datasets and quantitative experiments demonstrate that the proposed method can effectively extract complete line segments from input images compared to other advanced line segment detectors. The implementation is available at https://github.com/zyl0609/AAGLSD.

cs.CV↗

arXiv2025

Bootstrapping SU(3) Lattice Yang-Mills Theory

We apply the positivity bootstrap approach to SU(3) lattice Yang-Mills (YM) theory, extending previous studies of large N and SU(2) theories by incorporating multiple-trace Wilson loop operators. By utilizing Hermitian and reflection positivity conditions, alongside Schwinger-Dyson (SD) loop equations, we compute rigorous bounds for the expectation values of plaquette Wilson loops in 2D, 3D, and 4D YM theories. Our results exhibit clear convergence and are consistent with known analytic or numerical results. To enhance the approach, we introduce a novel twist-reflection positivity condition, which we prove to be exact in 2D YM theory. Additionally, we propose a dimensional-reduction truncation, where Wilson loop operators are effectively restricted to a lower-dimensional subplane, significantly simplifying computations. SD equations for double-trace Wilson loops are also derived in detail. Our findings suggest that the positivity bootstrap method is broadly applicable to higher-rank gauge theories beyond single-trace cases, providing a solid foundation for further non-perturbative investigations of gauge theories using positivity-based methods.

hep-th↗

arXiv2025

PixelatedScatter: Arbitrary-level Visual Abstraction for Large-scale Multiclass Scatterplots

Overdraw is inevitable in large-scale scatterplots. Current scatterplot abstraction methods lose features in medium-to-low density regions. We propose a visual abstraction method designed to provide better feature preservation across arbitrary abstraction levels for large-scale scatterplots, particularly in medium-to-low density regions. The method consists of three closely interconnected steps: first, we partition the scatterplot into iso-density regions and equalize visual density; then, we allocate pixels for different classes within each region; finally, we reconstruct the data distribution based on pixels. User studies, quantitative and qualitative evaluations demonstrate that, compared to previous methods, our approach better preserves features and exhibits a special advantage when handling ultra-high dynamic range data distributions.

cs.MM↗

arXiv2025

The Data Fusion Labeler (dFL): Challenges and Solutions to Data Harmonization, Labeling, and Provenance in Fusion Energy

Fusion energy research increasingly depends on the ability to integrate heterogeneous, multimodal datasets from high-resolution diagnostics, control systems, and multiscale simulations. The sheer volume and complexity of these datasets demand the development of new tools capable of systematically harmonizing and extracting knowledge across diverse modalities. The Data Fusion Labeler (dFL) is introduced as a unified workflow instrument that performs uncertainty-aware data harmonization, schema-compliant data fusion, and provenance-rich manual and automated labeling at scale. By embedding alignment, normalization, and labeling within a reproducible, operator-order-aware framework, dFL reduces time-to-analysis by greater than 50X (e.g., enabling >200 shots/hour to be consistently labeled rather than a handful per day), enhances label (and subsequently training) quality, and enables cross-device comparability. Case studies from DIII-D demonstrate its application to automated ELM detection and confinement regime classification, illustrating its potential as a core component of data-driven discovery, model validation, and real-time control in future burning plasma devices.

physics.plasm-ph↗

arXiv2025

Reasoning Language Model Inference Serving Unveiled: An Empirical Study

The reasoning large language model (RLLM) has been proven competitive in solving complex reasoning tasks such as mathematics, coding, compared to general LLM. However, the serving performance and behavior of RLLM remains unexplored, which may undermine the deployment and utilization of RLLM in real-world scenario. To close this gap, in this paper, we conduct a comprehensive study of RLLM service. We first perform a pilot study on comparing the serving performance between RLLM and traditional LLM and reveal that there are several distinct differences regarding serving behavior: (1) significant memory usage and fluctuations; (2) straggler requests; (3) adaptive running time; (4) domain preference. Then we further investigate whether existing inference optimization techniques are valid for RLLM. Our main takeaways are that model quantization methods and speculative decoding can improve service system efficiency with small compromise to RLLM accuracy, while prefix caching, KV cache quantization may even degrade accuracy or serving performance for small RLLM. Lastly, we conduct evaluation under real world workload modeled by Gamma distribution to verify our findings. Empirical results of real world workload evaluation across different dataset are aligned with our main findings regarding RLLM serving. We hope our work can provide the research community and industry with insights to advance RLLM inference serving.

cs.LG↗

arXiv2025

AnTKV: Anchor Token-Aware Sub-Bit Vector Quantization for KV Cache in Large Language Models

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.

cs.CL↗

arXiv2025

ChunkKV: Semantic-Preserving KV Cache Compression for Efficient Long-Context LLM Inference

Large Language Models (LLMs) require significant GPU memory when processing long texts, with the key value (KV) cache consuming up to 70\% of total memory during inference. Although existing compression methods reduce memory by evaluating the importance of individual tokens, they overlook critical semantic relationships between tokens, resulting in fragmented context and degraded performance. We introduce ChunkKV, which fundamentally reimagines KV cache compression by treating semantic chunks - rather than isolated tokens - as basic compression units. This approach preserves complete linguistic structures and contextual integrity, ensuring that essential meaning is retained even under aggressive compression. Our innovation includes a novel layer-wise index reuse technique that exploits the higher cross-layer similarity of preserved indices in ChunkKV, reducing computational overhead and improving throughput by 26.5\%. Comprehensive evaluations on challenging benchmarks: LongBench, Needle-In-A-HayStack, GSM8K, and JailbreakV demonstrate that ChunkKV outperforms state-of-the-art methods by up to 8.7\% in precision while maintaining the same compression ratio. These results confirm that semantic-aware compression significantly enhances both efficiency and performance for long-context LLM inference, providing a simple yet effective solution to the memory bottleneck problem. The code is available at \href{https://github.com/NVIDIA/kvpress}{link}.

cs.CL↗

arXiv2025

VeritasFi: An Adaptable, Multi-tiered RAG Framework for Multi-modal Financial Question Answering

Retrieval-Augmented Generation (RAG) is becoming increasingly essential for Question Answering (QA) in the financial sector, where accurate and contextually grounded insights from complex public disclosures are crucial. However, existing financial RAG systems face two significant challenges: (1) they struggle to process heterogeneous data formats, such as text, tables, and figures; and (2) they encounter difficulties in balancing general-domain applicability with company-specific adaptation. To overcome these challenges, we present VeritasFi, an innovative hybrid RAG framework that incorporates a multi-modal preprocessing pipeline alongside a cutting-edge two-stage training strategy for its re-ranking component. VeritasFi enhances financial QA through three key innovations: (1) A multi-modal preprocessing pipeline that seamlessly transforms heterogeneous data into a coherent, machine-readable format. (2) A tripartite hybrid retrieval engine that operates in parallel, combining deep multi-path retrieval over a semantically indexed document corpus, real-time data acquisition through tool utilization, and an expert-curated memory bank for high-frequency questions, ensuring comprehensive scope, accuracy, and efficiency. (3) A two-stage training strategy for the document re-ranker, which initially constructs a general, domain-specific model using anonymized data, followed by rapid fine-tuning on company-specific data for targeted applications. By integrating our proposed designs, VeritasFi presents a groundbreaking framework that greatly enhances the adaptability and robustness of financial RAG systems, providing a scalable solution for both general-domain and company-specific QA tasks. Code accompanying this work is available at https://github.com/simplew4y/VeritasFi.git.

cs.IR↗