arXiv subjects

Pengyu Li

Publications and source records attributed to Pengyu Li.

At least 19 recordsLinked to original sources

arXiv2026

OCGQuant: Outlier-Companion Grouping for NVFP4 Quantization

NVFP4 is an efficient microscaling format for low-bit inference, but activation outliers can still degrade quantization accuracy within NVFP4 blocks. Within each quantization block, large activations can dominate the block scale, increasing the quantization error of the remaining values sharing the same scale. Existing post-training quantization (PTQ) methods mitigate outlier errors through strategies such as mixed precision, rotation, or residual compensation, but these approaches are either not specifically tailored to NVFP4 or introduce additional computation. In this work, we revisit NVFP4 from a channel-grouping perspective and define the reducible error incurred by remaining block values under the scale set by the block maximum as Collateral Quantization Error. Based on this insight, we propose OCGQuant, a post-training quantization method centered on Outlier-Companion Grouping (OCG), which adaptively pairs outlier channels with low-magnitude companion channels to improve NVFP4 activation block composition. Experiments on Llama3 and Qwen3 show that OCGQuant achieves the lowest WikiText-2 perplexity and highest average downstream accuracy among evaluated PTQ methods, while maintaining prefill speedup close to RTN and matching its peak decoding memory. Code is available at https://github.com/Eshamont/OCGQuant.

cs.CL↗

arXiv2026

GLaQ: Grounding Latent Queries in Visual Evidence for Multimodal Reasoning

Chain-of-thought reasoning has substantially improved the problem-solving capabilities of multimodal large language models. Fine-grained visual evidence, however, remains difficult to preserve and reuse across text-based reasoning steps. To address this limitation, tool-augmented thinking-with-images methods maintain visual access externally by revisiting or manipulating the image, but require predefined tools and additional inference-time processing. As an internal alternative, continuous visual latent reasoning retains intermediate computation in hidden states. However, its prevailing autoregressive construction makes each latent state depend on its predecessors, so later states may repeat information already present in the latent sequence rather than capture complementary visual details. We introduce GLaQ, a grounded latent-query framework that replaces sequential latent rollout with a fixed set of context-conditioned queries grounded in the original visual tokens. The grounded queries are reinjected for answer generation, providing direct and coordinated access to source visual evidence. We train GLaQ with localized-view supervision followed by reinforcement learning under task-level rewards. Across five benchmarks for fine-grained visual understanding and perception, GLaQ-7B gains 5.99--9.66\% over its base model and leads all compared visual latent methods, suggesting that direct query-to-image grounding can recover localized evidence from the full image without external visual operations or autoregressive latent rollouts.

cs.CV↗

arXiv2026

QUMem: Personalized Memory for Query-Conditioned User-State Inference in LLM Agents

Large language model (LLM) agents increasingly use external memory systems to support personalization by drawing on long and evolving interaction histories, in which user preferences may be distributed across time, change with context, and conflict with earlier evidence. However, existing systems face three limitations: fixed-turn, fixed-token, or session-based boundaries can mix unrelated dialogue or split an event from its causes, decisions, and outcomes; storing multiple pieces of user information from the same interaction as a single memory binds together items that serve different functions and should be independently retrievable; and treating the current task as a single top-$k$ retrieval query can return fragments that are individually relevant but fail to jointly capture preference evolution, temporal validity, and contextual applicability. We introduce \textsc{QUMem}, a structured memory framework for query-conditioned user-state inference. \textsc{QUMem} first segments interaction histories into variable-length episodes according to semantic continuity, then decomposes each episode into independently retrievable factual, preference, and transferable insight memories while preserving temporal positions and source evidence. At inference time, three sequential agents identify task-specific information needs, plan multi-query retrieval over the typed memory stores, and jointly infer a temporally and contextually valid user state for downstream response generation. \textsc{QUMem} achieves state-of-the-art performance on both PersonaMem and KnowU-Bench, demonstrating the effectiveness of query-conditioned user-state inference for long-term personalization.

cs.CL↗

arXiv2026

Post-Learning Inference for Combinatorial Optimizers with High-Dimensional Sparse Contextual Information via Minimal Directional Perturbation

We study post-learning inference for structural properties of data-dependent combinatorial optimizers. The target is whether an oracle optimizer, rather than a latent parameter or smooth functional, belongs to a prescribed class, such as a category-mix, inventory, or resource-feasibility class. We focus on a high-dimensional contextual multinomial logit model with sequentially adaptive data collection, where the parameter-to-optimizer map is discontinuous and the policy induces temporal dependence. We propose a novel perturbation test based on a nonsmooth max-difference revenue statistic comparing the best null assortment with the best alternative assortment. The test perturbs the estimated terminal revenue surface on the selected support: random unit directions capture directional uncertainty, while the minimal perturbation radius captures magnitude uncertainty and yields a p-value. This localizes inference near the null--alternative boundary and avoids uniform error control over the full candidate class. The data are collected by an $\ell_1$-penalized online likelihood policy that performs variable selection while controlling regret. Using a new anti-concentration argument for Gaussian maxima differences and martingale Gaussian coupling, we establish uniform estimation rates, effective support recovery, and asymptotic validity of the proposed p-value under adaptive assortment selection. We prove asymptotic size control and power consistency under a localized signal condition.

math.ST↗

arXiv2026

Visual-OPSD: Cross-Modal On-Policy Self-Distillation for Efficient Unified Multimodal Reasoning

Unified multimodal models (UMMs) interleave generated ''visual thoughts'' (VTs) with text reasoning to improve spatial tasks. This incurs roughly an order-of-magnitude inference cost from multi-step diffusion. We find this cost yields limited direct benefit. On ThinkMorph, removing or noising VTs barely changes accuracy across nine benchmarks. Once rendered, attention concentrates on the VT regardless of content. Yet a KL diagnostic shows that conditioning on a privileged VT trace shifts the model's completion distribution. This suggests the generation pathway encodes useful reasoning beyond the rendered pixels. Motivated by this gap, we propose Visual On-Policy Self-Distillation(Visual-OPSD). Teacher and student share identical weights but differ in context: the teacher sees privileged VTs while the student sees only the question. Token-level JSD distillation on on-policy student trajectories transfers the teacher's reasoning to a text-only student. Across nine benchmarks, Visual-OPSD improves over its generative teacher by $+3.40$pp with $14.3\times$ speedup (10.0s vs. 142.8s per sample) and outperforms same-scale VLMs by $+63.83$pp on VSP. A Gaussian-noise control ($+0.40$pp vs. $+10.28$pp for real VTs) and $58.4\%$ closure of the KL gap confirm that gains come from the semantic content of the generation pathway.

cs.CV↗

arXiv2026

Finding Kissing Numbers with Game-theoretic Reinforcement Learning

Since Isaac Newton first studied the Kissing Number Problem in 1694, determining the maximal number of non-overlapping spheres around a central sphere has remained a defining challenge in discrete geometry. As the local analogue of Hilbert's 18th problem, it has profound implications across geometry, number theory and information theory. Although lattices and codes have achieved significant progress, the field is confined to isolated extremal configurations, leaving underlying geometric principles obscured. Here we shift the object to the broader extremal configuration space, thereby opening a new path for the Kissing Number Problem. Accordingly, we recast this problem as a cooperative matrix-completion game, and train a reinforcement learning system, PackingStar, to solve it. One player fills cosine entries while the other corrects suboptimal ones, making explosive geometric complexity tractable. Working within extremal configuration spaces, PackingStar discovers new interpretable geometric structures that improve 15 strong bounds held for decades in kissing numbers and their generalizations, several of them provably optimal under natural inner products. These findings reveal the first explicit spherical-code realization of the Fischer group Fi22, extend the classical Euclidean representation of subgroup structure, and directly inspire subsequent breakthroughs by mathematicians. Overall, the work provides an early example of AI-driven progress on a Hilbert-calibre problem, showing how reinforcement learning advances mathematical discovery by unlocking more expressive objects.

cs.LG↗

arXiv2026

PathFinder: Advancing Path Loss Prediction for Single-to-Multi-Transmitter Scenario

Radio path loss prediction (RPP) is critical for optimizing 5G networks and enabling IoT, smart city, and similar applications. However, current deep learning-based RPP methods lack proactive environmental modeling, struggle with realistic multi-transmitter scenarios, and generalize poorly under distribution shifts, particularly when training/testing environments differ in building density or transmitter configurations. This paper identifies three key issues: (1) passive environmental modeling that overlooks transmitters and key environmental features; (2) overemphasis on single-transmitter scenarios despite real-world multi-transmitter prevalence; (3) excessive focus on in-distribution performance while neglecting distribution shift challenges. To address these, we propose PathFinder, a novel architecture that actively models buildings and transmitters via disentangled feature encoding and integrates Mask-Guided Low-Rank Attention to independently focus on receiver and building regions. We also introduce a Transmitter-Oriented Mixup strategy for robust training and a new benchmark, single-to-multi-transmitter RPP (S2MT-RPP), tailored to evaluate extrapolation performance (multi-transmitter testing after single-transmitter training). Experimental results show PathFinder outperforms state-of-the-art methods significantly, especially in challenging multi-transmitter scenarios. Our code and project site are available at: https://emorzz1g.github.io/PathFinder/.

cs.LG↗

arXiv2026

LogicGraph : Benchmarking Multi-Path Logical Reasoning via Neuro-Symbolic Generation and Verification

Evaluations of large language models (LLMs) primarily emphasize convergent logical reasoning, where success is defined by producing a single correct proof. However, many real-world reasoning problems admit multiple valid derivations, requiring models to explore diverse logical paths rather than committing to one route. To address this limitation, we introduce LogicGraph, the first benchmark aimed to systematically evaluate multi-path logical reasoning, constructed via a neuro-symbolic framework that leverages backward logic generation and semantic instantiation. This pipeline yields solver-verified reasoning problems formalized by high-depth multi-path reasoning and inherent logical distractions, where each instance is associated with an exhaustive set of minimal proofs. We further propose a reference-free evaluation framework to rigorously assess model performance in both convergent and divergent regimes. Experiments on state-of-the-art language models reveal a common limitation: models tend to commit early to a single route and fail to explore alternatives, and the coverage gap grows substantially with reasoning depth. LogicGraph exposes this divergence gap and provides actionable insights to motivate future improvements. Our code and data will be released at https://github.com/kkkkarry/LogicGraph.

cs.AI↗

arXiv2026

AERO: Autonomous Evolutionary Reasoning Optimization via Endogenous Dual-Loop Feedback

Large Language Models (LLMs) have achieved significant success in complex reasoning but remain bottlenecked by reliance on expert-annotated data and external verifiers. While existing self-evolution paradigms aim to bypass these constraints, they often fail to identify the optimal learning zone and risk reinforcing collective hallucinations and incorrect priors through flawed internal feedback. To address these challenges, we propose \underline{A}utonomous \underline{E}volutionary \underline{R}easoning \underline{O}ptimization (AERO), an unsupervised framework that achieves autonomous reasoning evolution by internalizing self-questioning, answering, and criticism within a synergistic dual-loop system. Inspired by the \textit{Zone of Proximal Development (ZPD)} theory, AERO utilizes entropy-based positioning to target the ``solvability gap'' and employs Independent Counterfactual Correction for robust verification. Furthermore, we introduce a Staggered Training Strategy to synchronize capability growth across functional roles and prevent curriculum collapse. Extensive evaluations across nine benchmarks spanning three domains demonstrate that AERO achieves average performance improvements of 4.57\% on Qwen3-4B-Base and 5.10\% on Qwen3-8B-Base, outperforming competitive baselines. Code is available at https://github.com/mira-ai-lab/AERO.

cs.CL↗

arXiv2026

$\textbf{AGT$^{AO}$}$: Robust and Stabilized LLM Unlearning via Adversarial Gating Training with Adaptive Orthogonality

While Large Language Models (LLMs) have achieved remarkable capabilities, they unintentionally memorize sensitive data, posing critical privacy and security risks. Machine unlearning is pivotal for mitigating these risks, yet existing paradigms face a fundamental dilemma: aggressive unlearning often induces catastrophic forgetting that degrades model utility, whereas conservative strategies risk superficial forgetting, leaving models vulnerable to adversarial recovery. To address this trade-off, we propose $\textbf{AGT$^{AO}$}$ (Adversarial Gating Training with Adaptive Orthogonality), a unified framework designed to reconcile robust erasure with utility preservation. Specifically, our approach introduces $\textbf{Adaptive Orthogonality (AO)}$ to dynamically mitigate geometric gradient conflicts between forgetting and retention objectives, thereby minimizing unintended knowledge degradation. Concurrently, $\textbf{Adversarial Gating Training (AGT)}$ formulates unlearning as a latent-space min-max game, employing a curriculum-based gating mechanism to simulate and counter internal recovery attempts. Extensive experiments demonstrate that $\textbf{AGT$^{AO}$}$ achieves a superior trade-off between unlearning efficacy (KUR $\approx$ 0.01) and model utility (MMLU 58.30). Code is available at https://github.com/TiezMind/AGT-unlearning.

cs.LG↗

arXiv2025

Suppression and Deterrence: Revisiting the Welfare Consequence of HIV Stigma

The government's effort to alleviate HIV stigma has been justified by the suppression effect of stigma on the HIV testing rate. Nevertheless, the deterrence effect of stigma on undesirable sexual behaviours has long been overlooked. This study adapts the existing framework on HIV stigma with an additional stage that formally models people's choices on whether to take preventive measures in sex. The model shows that, when sex is explicitly modelled, the suppression and deterrence effects coexist, which makes the net societal impact of HIV stigma ambiguous. A utilitarian welfare analysis concludes that the welfare-maximizing stigma level can be higher than its natural level, implying that the government's effort to reduce stigma is not always welfare-improving. Instead, the study provides a rationale for maintaining a certain level of HIV stigma to maximize social welfare.

econ.GN↗

arXiv2025

An RGB-D Image Dataset for Lychee Detection and Maturity Classification for Robotic Harvesting

Lychee is a high-value subtropical fruit. The adoption of vision-based harvesting robots can significantly improve productivity while reduce reliance on labor. High-quality data are essential for developing such harvesting robots. However, there are currently no consistently and comprehensively annotated open-source lychee datasets featuring fruits in natural growing environments. To address this, we constructed a dataset to facilitate lychee detection and maturity classification. Color (RGB) images were acquired under diverse weather conditions, and at different times of the day, across multiple lychee varieties, such as Nuomici, Feizixiao, Heiye, and Huaizhi. The dataset encompasses three different ripeness stages and contains 11,414 images, consisting of 878 raw RGB images, 8,780 augmented RGB images, and 1,756 depth images. The images are annotated with 9,658 pairs of lables for lychee detection and maturity classification. To improve annotation consistency, three individuals independently labeled the data, and their results were then aggregated and verified by a fourth reviewer. Detailed statistical analyses were done to examine the dataset. Finally, we performed experiments using three representative deep learning models to evaluate the dataset. It is publicly available for academic

cs.CV↗

arXiv2025

HunyuanVideo: A Systematic Framework For Large Video Generative Models

Recent advancements in video generation have significantly impacted daily life for both individuals and industries. However, the leading video generation models remain closed-source, resulting in a notable performance gap between industry capabilities and those available to the public. In this report, we introduce HunyuanVideo, an innovative open-source video foundation model that demonstrates performance in video generation comparable to, or even surpassing, that of leading closed-source models. HunyuanVideo encompasses a comprehensive framework that integrates several key elements, including data curation, advanced architectural design, progressive model scaling and training, and an efficient infrastructure tailored for large-scale model training and inference. As a result, we successfully trained a video generative model with over 13 billion parameters, making it the largest among all open-source models. We conducted extensive experiments and implemented a series of targeted designs to ensure high visual quality, motion dynamics, text-video alignment, and advanced filming techniques. According to evaluations by professionals, HunyuanVideo outperforms previous state-of-the-art models, including Runway Gen-3, Luma 1.6, and three top-performing Chinese video generative models. By releasing the code for the foundation model and its applications, we aim to bridge the gap between closed-source and open-source communities. This initiative will empower individuals within the community to experiment with their ideas, fostering a more dynamic and vibrant video generation ecosystem. The code is publicly available at https://github.com/Tencent/HunyuanVideo.

cs.CV↗

arXiv2025

LarQucut: A New Cutting and Mapping Approach for Large-sized Quantum Circuits in Distributed Quantum Computing (DQC) Environments

Distributed quantum computing (DQC) is a promising way to achieve large-scale quantum computing. However, mapping large-sized quantum circuits in DQC is a challenging job; for example, it is difficult to find an ideal cutting and mapping solution when many qubits, complicated qubit operations, and diverse QPUs are involved. In this study, we propose LarQucut, a new quantum circuit cutting and mapping approach for large-sized circuits in DQC. LarQucut has several new designs. (1) LarQucut can have cutting solutions that use fewer cuts, and it does not cut a circuit into independent sub-circuits, therefore reducing the overall cutting and computing overheads. (2) LarQucut finds isomorphic sub-circuits and reuses their execution results. So, LarQucut can reduce the number of sub-circuits that need to be executed to reconstruct the large circuit's output, reducing the time spent on sampling the sub-circuits. (3) We design an adaptive quantum circuit mapping approach, which identifies qubit interaction patterns and accordingly enables the best-fit mapping policy in DQC. The experimental results show that, for large circuits with hundreds to thousands of qubits in DQC, LarQucut can provide a better cutting and mapping solution with lower overall overheads and achieves results closer to the ground truth.

quant-ph↗

arXiv2024

A New Perspective on Time Series Anomaly Detection: Faster Patch-based Broad Learning System

Time series anomaly detection (TSAD) has been a research hotspot in both academia and industry in recent years. Deep learning methods have become the mainstream research direction due to their excellent performance. However, new viewpoints have emerged in recent TSAD research. Deep learning is not required for TSAD due to limitations such as slow deep learning speed. The Broad Learning System (BLS) is a shallow network framework that benefits from its ease of optimization and speed. It has been shown to outperform machine learning approaches while remaining competitive with deep learning. Based on the current situation of TSAD, we propose the Contrastive Patch-based Broad Learning System (CPatchBLS). This is a new exploration of patching technique and BLS, providing a new perspective for TSAD. We construct Dual-PatchBLS as a base through patching and Simple Kernel Perturbation (SKP) and utilize contrastive learning to capture the differences between normal and abnormal data under different representations. To compensate for the temporal semantic loss caused by various patching, we propose CPatchBLS with model level integration, which takes advantage of BLS's fast feature to build model-level integration and improve model detection. Using five real-world series anomaly detection datasets, we confirmed the method's efficacy, outperforming previous deep learning and machine learning methods while retaining a high level of computing efficiency.

cs.LG↗

arXiv2024

EMDFNet: Efficient Multi-scale and Diverse Feature Network for Traffic Sign Detection

The detection of small objects, particularly traffic signs, is a critical subtask within object detection and autonomous driving. Despite the notable achievements in previous research, two primary challenges persist. Firstly, the main issue is the singleness of feature extraction. Secondly, the detection process fails to effectively integrate with objects of varying sizes or scales. These issues are also prevalent in generic object detection. Motivated by these challenges, in this paper, we propose a novel object detection network named Efficient Multi-scale and Diverse Feature Network (EMDFNet) for traffic sign detection that integrates an Augmented Shortcut Module and an Efficient Hybrid Encoder to address the aforementioned issues simultaneously. Specifically, the Augmented Shortcut Module utilizes multiple branches to integrate various spatial semantic information and channel semantic information, thereby enhancing feature diversity. The Efficient Hybrid Encoder utilizes global feature fusion and local feature interaction based on various features to generate distinctive classification features by integrating feature information in an adaptable manner. Extensive experiments on the Tsinghua-Tencent 100K (TT100K) benchmark and the German Traffic Sign Detection Benchmark (GTSDB) demonstrate that our EMDFNet outperforms other state-of-the-art detectors in performance while retaining the real-time processing capabilities of single-stage models. This substantiates the effectiveness of EMDFNet in detecting small traffic signs.

cs.CV↗

arXiv2024

Sim2Real in Reconstructive Spectroscopy: Deep Learning with Augmented Device-Informed Data Simulation

This work proposes a deep learning (DL)-based framework, namely Sim2Real, for spectral signal reconstruction in reconstructive spectroscopy, focusing on efficient data sampling and fast inference time. The work focuses on the challenge of reconstructing real-world spectral signals under the extreme setting where only device-informed simulated data are available for training. Such device-informed simulated data are much easier to collect than real-world data but exhibit large distribution shifts from their real-world counterparts. To leverage such simulated data effectively, a hierarchical data augmentation strategy is introduced to mitigate the adverse effects of this domain shift, and a corresponding neural network for the spectral signal reconstruction with our augmented data is designed. Experiments using a real dataset measured from our spectrometer device demonstrate that Sim2Real achieves significant speed-up during the inference while attaining on-par performance with the state-of-the-art optimization-based methods.

cs.LG↗

arXiv2024

PersonificationNet: Making customized subject act like a person

Recently customized generation has significant potential, which uses as few as 3-5 user-provided images to train a model to synthesize new images of a specified subject. Though subsequent applications enhance the flexibility and diversity of customized generation, fine-grained control over the given subject acting like the person's pose is still lack of study. In this paper, we propose a PersonificationNet, which can control the specified subject such as a cartoon character or plush toy to act the same pose as a given referenced person's image. It contains a customized branch, a pose condition branch and a structure alignment module. Specifically, first, the customized branch mimics specified subject appearance. Second, the pose condition branch transfers the body structure information from the human to variant instances. Last, the structure alignment module bridges the structure gap between human and specified subject in the inference stage. Experimental results show our proposed PersonificationNet outperforms the state-of-the-art methods.

cs.CV↗