arXiv subjects

Zhenyu Li

Publications and source records attributed to Zhenyu Li.

At least 55 records · Page 3Linked to original sources

arXiv2026

SpatiaLoc: Leveraging Multi-Level Spatial Enhanced Descriptors for Cross-Modal Localization

Cross-modal localization using text and point clouds enables robots to localize themselves via natural language descriptions, with applications in autonomous navigation and interaction between humans and robots. In this task, objects often recur across text and point clouds, making spatial relationships the most discriminative cues for localization. Given this characteristic, we present SpatiaLoc, a framework utilizing a coarse-to-fine strategy that emphasizes spatial relationships at both the instance and global levels. In the coarse stage, we introduce a Bezier Enhanced Object Spatial Encoder (BEOSE) that models spatial relationships at the instance level using quadratic Bezier curves. Additionally, a Frequency Aware Encoder (FAE) generates spatial representations in the frequency domain at the global level. In the fine stage, an Uncertainty Aware Gaussian Fine Localizer (UGFL) regresses 2D positions by modeling predictions as Gaussian distributions with a loss function aware of uncertainty. Extensive experiments on KITTI360Pose demonstrate that SpatiaLoc significantly outperforms existing state-of-the-art (SOTA) methods.

cs.CV↗

arXiv2025

Depth Anything 3: Recovering the Visual Space from Any Views

We present Depth Anything 3 (DA3), a model that predicts spatially consistent geometry from an arbitrary number of visual inputs, with or without known camera poses. In pursuit of minimal modeling, DA3 yields two key insights: a single plain transformer (e.g., vanilla DINO encoder) is sufficient as a backbone without architectural specialization, and a singular depth-ray prediction target obviates the need for complex multi-task learning. Through our teacher-student training paradigm, the model achieves a level of detail and generalization on par with Depth Anything 2 (DA2). We establish a new visual geometry benchmark covering camera pose estimation, any-view geometry and visual rendering. On this benchmark, DA3 sets a new state-of-the-art across all tasks, surpassing prior SOTA VGGT by an average of 44.3% in camera pose accuracy and 25.1% in geometric accuracy. Moreover, it outperforms DA2 in monocular depth estimation. All models are trained exclusively on public academic datasets.

cs.CV↗

arXiv2025

Quantum-centric machine learning for molecular dynamics

Accurate and efficient prediction of electronic wavefunctions is central to ab initio molecular dynamics (AIMD) and electronic structure theory. However, conventional ab initio methods require self-consistent optimization of electronic states at every nuclear configuration, leading to prohibitive computational costs, especially for large or strongly correlated systems. Here, we introduce a quantum-centric machine learning (QCML) model-a hybrid quantum-classical framework that integrates parameterized quantum circuits (PQCs) with Transformer-based machine learning to directly predict molecular wavefunctions and quantum observables. By pretraining the Transformer on a diverse dataset of molecules and ansatz types and subsequently fine-tuning it for specific systems, QCML learns transferable mappings between molecular descriptors and PQC parameters, eliminating the need for iterative variational optimization. The pretrained model achieves chemical accuracy in potential energy surfaces, atomic forces, and dipole moments across multiple molecules and ansatzes, and enables efficient AIMD simulations with infrared spectra prediction. This work establishes a scalable and transferable quantum-centric machine learning paradigm, bridging variational quantum algorithms and modern deep learning for next-generation molecular simulation and quantum chemistry applications.

physics.chem-ph↗

arXiv2025

A step toward Chen-Lih-Wu conjecture

An equitable $k$-coloring of a graph is a proper $k$-coloring where the sizes of any two different color classes differ by at most one. In 1973, Meyer conjectured that every connected graph $G$ has an equitable $k$-coloring for some $k\leq Δ(G)$, unless $G$ is a complete graph or an odd cycle. Chen, Lih, and Wu strengthened this in 1994 by conjecturing that for $k\geq 3$, the only connected graphs of maximum degree at most $k$ with no equitable $k$-coloring are the complete bipartite graph $K_{k,k}$ for odd $k$ and the complete graph $K_{k+1}$. A more refined conjecture was proposed by Kierstead and Kostochka, relaxing the maximum degree condition to an Ore-type condition. Their conjecture states the following: for $k\geq 3$, if $G$ is an $n$-vertex graph such that $d(x) + d(y)\leq 2k$ for every edge $xy\in E(G)$, and $G$ admits no equitable $k$-coloring, then $G$ contains either $K_{k+1}$ or $K_{m,2k-m}$ for some odd $m$. We prove that for any constant $c>0$ and all sufficiently large $n$, the latter two conjectures hold for every $k\geq cn$. Our proof yields an algorithm with polynomial time that decides whether $G$ has an equitable $k$-coloring, thereby answering a conjecture of Kierstead, Kostochka, Mydlarz, and Szemerédi when $k \ge cn$.

math.CO↗

arXiv2025

XIFBench: Evaluating Large Language Models on Multilingual Instruction Following

Large Language Models (LLMs) have demonstrated remarkable instruction-following capabilities across various applications. However, their performance in multilingual settings lacks systematic investigation, with existing evaluations lacking fine-grained constraint analysis across diverse linguistic contexts. We introduce XIFBench, a comprehensive constraint-based benchmark for evaluating multilingual instruction-following abilities of LLMs, comprising 558 instructions with 0-5 additional constraints across five categories (Content, Style, Situation, Format, and Numerical) in six languages spanning different resource levels. To support reliable and consistent cross-lingual evaluation, we implement three methodological innovations: cultural accessibility annotation, constraint-level translation validation, and requirement-based evaluation using English requirements as semantic anchors across languages. Extensive experiments with various LLMs not only quantify performance disparities across resource levels but also provide detailed insights into how language resources, constraint categories, instruction complexity, and cultural specificity influence multilingual instruction-following. Our code and data are available at https://github.com/zhenyuli801/XIFBench.

cs.CL↗

arXiv2025

SUMMA: A Multimodal Large Language Model for Advertisement Summarization

Understanding multimodal video ads is crucial for improving query-ad matching and relevance ranking on short video platforms, enhancing advertising effectiveness and user experience. However, the effective utilization of multimodal information with high commercial value still largely constrained by reliance on highly compressed video embeddings-has long been inadequate. To address this, we propose SUMMA (the abbreviation of Summarizing MultiModal Ads), a multimodal model that automatically processes video ads into summaries highlighting the content of highest commercial value, thus improving their comprehension and ranking in Douyin search-advertising systems. SUMMA is developed via a two-stage training strategy-multimodal supervised fine-tuning followed by reinforcement learning with a mixed reward mechanism-on domain-specific data containing video frames and ASR/OCR transcripts, generating commercially valuable and explainable summaries. We integrate SUMMA-generated summaries into our production pipeline, directly enhancing the candidate retrieval and relevance ranking stages in real search-advertising systems. Both offline and online experiments show substantial improvements over baselines, with online results indicating a statistically significant 1.5% increase in advertising revenue. Our work establishes a novel paradigm for condensing multimodal information into representative texts, effectively aligning visual ad content with user query intent in retrieval and recommendation scenarios.

cs.IR↗

arXiv2025

Pinpointing crucial steps: Attribution-based Credit Assignment for Verifiable Reinforcement Learning

While Reinforcement Learning with Verifiable Rewards (RLVR) enhances complex reasoning in LLMs, current methods struggle to balance exploration and exploitation. This leads to critical issues like inaccurate credit assignment for intermediate steps and premature entropy collapse, limiting model performance. To address this, we introduce Attribution-based Contribution to Policy Optimization (ACPO), a phased framework that incorporates a difficulty-aware curriculum. ACPO improves exploration by using trajectory semantic segmentation and an attribution-based representation to dynamically regulate policy entropy, thus mitigating its collapse. Concurrently, it enhances exploitation with a factorized reward system that precisely quantifies the hierarchical contribution of each reasoning step, ensuring accurate credit assignment. Extensive experiments on challenging benchmarks, including AIME, MATH, and AMC, demonstrate that ACPO significantly outperforms existing state-of-the-art approaches.

cs.LG↗

arXiv2025

Scalable tensor network algorithm for quantum impurity problems

The Grassmann time-evolving matrix product operator method has shown great potential as a general-purpose quantum impurity solver, as its numerical errors can be well-controlled and it is flexible to be applied on both the imaginary- and real-time axis. However, a major limitation of it is that its computational cost grows exponentially with the number of impurity flavors. In this work, we propose a multi-flavor extension of it to overcome this limitation. The key insight is that to calculate multi-time correlation functions on one or a few impurity flavors, one could integrate out the degrees of freedom of the rest flavors before hand, which could greatly simplify the calculation. The idea is particularly effective for quantum impurity problems with diagonal hybridization function, i.e., each impurity flavor is coupled to an independent bath, a setting which is commonly used in the field. We demonstrate the accuracy and scalability of our method for the imaginary time evolution of impurity problems with up to three impurity orbitals, i.e., 6 flavors, and benchmark our results against continuous-time quantum Monte Carlo calculations. Our method paves the way of scaling up tensor network algorithms to solve large-scale quantum impurity problems.

cond-mat.str-el↗

arXiv2025

Vehicle-Scene Interaction: A Text-Driven 3D Lidar Place Recognition Method for Autonomous Driving

Environment description-based localization in large-scale point cloud maps constructed through remote sensing is critically significant for the advancement of large-scale autonomous systems, such as delivery robots operating in the last mile. However, current approaches encounter challenges due to the inability of point cloud encoders to effectively capture local details and long-range spatial relationships, as well as a significant modality gap between text and point cloud representations. To address these challenges, we present Des4Pos, a novel two-stage text-driven remote sensing localization framework. In the coarse stage, the point-cloud encoder utilizes the Multi-scale Fusion Attention Mechanism (MFAM) to enhance local geometric features, followed by a bidirectional Long Short-Term Memory (LSTM) module to strengthen global spatial relationships. Concurrently, the Stepped Text Encoder (STE) integrates cross-modal prior knowledge from CLIP [1] and aligns text and point-cloud features using this prior knowledge, effectively bridging modality discrepancies. In the fine stage, we introduce a Cascaded Residual Attention (CRA) module to fuse cross-modal features and predict relative localization offsets, thereby achieving greater localization precision. Experiments on the KITTI360Pose test set demonstrate that Des4Pos achieves state-of-the-art performance in text-to-point-cloud place recognition. Specifically, it attains a top-1 accuracy of 40% and a top-10 accuracy of 77% under a 5-meter radius threshold, surpassing the best existing methods by 7% and 7%, respectively.

cs.CV↗

arXiv2025

Scalable parallel simulation of quantum circuits on CPU and GPU systems

Quantum computing enables parallelism through superposition and entanglement and offers advantages over classical computing architectures. However, due to the limitations of current quantum hardware in the noisy intermediate-scale quantum (NISQ) era, classical simulation remains a critical tool for developing quantum algorithms. In this research, we present a comprehensive parallelization solution for the Q$^2$Chemistry software package, delivering significant performance improvements for the full-amplitude simulator on both CPU and GPU platforms. By incorporating batch-buffered overlap processing, dependency-aware gate contraction and staggered multi-gate parallelism, our optimizations significantly enhance the simulation speed compared to unoptimized baselines, demonstrating the effectiveness of hybrid-level parallelism in HPC systems. Benchmark results show that Q$^2$Chemistry consistently outperforms current state-of-the-art open-source simulators across various circuit types. These benchmarks highlight the capability of Q$^2$Chemistry to effectively handle large-scale quantum simulations with high efficiency and high portability.

quant-ph↗

arXiv2025

Negative Matters: Multi-Granularity Hard-Negative Synthesis and Anchor-Token-Aware Pooling for Enhanced Text Embeddings

Text embedding models are essential for various natural language processing tasks, enabling the effective encoding of semantic information into dense vector representations. These models are typically optimized using triplets of (query, positive, negative) data pairs for contrastive learning, where the negative samples play a critical role in enhancing the model's ability to discern subtle semantic distinctions. In this work, we introduce a Multi-Granularity Hard-negative (MGH) synthesis framework that leverages large language models (LLMs) to generate diverse negative samples with varying levels of similarity with the query. This approach facilitates a coarse-to-fine curriculum learning strategy during supervised training, allowing the embedding model to progressively learn more nuanced semantic representations. Meanwhile, we propose an Anchor Token Aware (ATA) pooling method that assigns higher weights to anchor tokens based on aggregation patterns observed in LLMs, improving text embedding accuracy without increasing model complexity. Comprehensive experiments on the MTEB benchmark demonstrate that our methods achieve state-of-the-art performance, surpassing existing synthesis strategies both with synthetic data and when combined with public retrieval datasets.

cs.CL↗

arXiv2025

Infinite Grassmann time-evolving matrix product operators for quantum impurity problems after a quench

An emergent numerical approach to solve quantum impurity problems is to encode the impurity path integral as a matrix product state. For time-dependent problems, the cost of this approach generally scales with the evolution time. Here we consider a common non-equilibrium scenario where an impurity, initially in equilibrium with a thermal bath, is driven out of equilibrium by a sudden quench of the impurity Hamiltonian. Despite that there is no time-translational invariance in the problem, we show that we could still make full use of the infinite matrix product state technique, resulting in a method whose cost is essentially independent of the evolution time. We demonstrate the effectiveness of this method in the integrable case against exact diagonalization, and against existing calculations on the L-shaped Kadanoff-Baym contour in the general case. Our method could be a very competitive method for studying long-time non-equilibrium quantum dynamics, and be potentially used as an efficient impurity solver in the non-equilibrium dynamical mean field theory.

cond-mat.str-el↗

arXiv2025

Maximum Score Routing For Mixture-of-Experts

Routing networks in sparsely activated mixture-of-experts (MoE) dynamically allocate input tokens to top-k experts through differentiable sparse transformations, enabling scalable model capacity while preserving computational efficiency. Traditional MoE networks impose an expert capacity constraint to ensure GPU-friendly computation. However, this leads to token dropping when capacity is saturated and results in low hardware efficiency due to padding in underutilized experts. Removing the capacity constraint, in turn, compromises load balancing and computational efficiency. To address these issues, we propose Maximum Score Routing ($\mathbf{MaxScore}$), a novel MoE routing paradigm that models routing as a minimum-cost maximum-flow problem and integrates a SoftTopk operator. MaxScore resolves the fundamental limitations of iterative rerouting and optimal transport formulations, achieving lower training losses and higher evaluation scores at equivalent FLOPs compared to both constrained and unconstrained baselines. Implementation details and experimental configurations can be obtained from $\href{https://github.com/dongbw18/MaxScore.git}{MaxScore}$.

cs.LG↗

arXiv2025

MyGO: Make your Goals Obvious, Avoiding Semantic Confusion in Prostate Cancer Lesion Region Segmentation

Early diagnosis and accurate identification of lesion location and progression in prostate cancer (PCa) are critical for assisting clinicians in formulating effective treatment strategies. However, due to the high semantic homogeneity between lesion and non-lesion areas, existing medical image segmentation methods often struggle to accurately comprehend lesion semantics, resulting in the problem of semantic confusion. To address this challenge, we propose a novel Pixel Anchor Module, which guides the model to discover a sparse set of feature anchors that serve to capture and interpret global contextual information. This mechanism enhances the model's nonlinear representation capacity and improves segmentation accuracy within lesion regions. Moreover, we design a self-attention-based Top_k selection strategy to further refine the identification of these feature anchors, and incorporate a focal loss function to mitigate class imbalance, thereby facilitating more precise semantic interpretation across diverse regions. Our method achieves state-of-the-art performance on the PI-CAI dataset, demonstrating 69.73% IoU and 74.32% Dice scores, and significantly improving prostate cancer lesion detection.

eess.IV↗

arXiv2025

BenchDepth: Are We on the Right Way to Evaluate Depth Foundation Models?

Depth estimation is a fundamental task in computer vision with diverse applications. Recent advancements in deep learning have led to powerful depth foundation models (DFMs), yet their evaluation remains challenging due to inconsistencies in existing protocols. Traditional benchmarks rely on alignment-based metrics that introduce biases, favor certain depth representations, and complicate fair comparisons. In this work, we propose BenchDepth, a new benchmark that evaluates DFMs through five carefully selected downstream proxy tasks: depth completion, stereo matching, monocular feed-forward 3D scene reconstruction, SLAM, and vision-language spatial understanding. Unlike conventional evaluation protocols, our approach assesses DFMs based on their practical utility in real-world applications, bypassing problematic alignment procedures. We benchmark eight state-of-the-art DFMs and provide an in-depth analysis of key findings and observations. We hope our work sparks further discussion in the community on best practices for depth model evaluation and paves the way for future research and advancements in depth estimation.

cs.CV↗

arXiv2025

OptiCorNet: Optimizing Sequence-Based Context Correlation for Visual Place Recognition

Visual Place Recognition (VPR) in dynamic and perceptually aliased environments remains a fundamental challenge for long-term localization. Existing deep learning-based solutions predominantly focus on single-frame embeddings, neglecting the temporal coherence present in image sequences. This paper presents OptiCorNet, a novel sequence modeling framework that unifies spatial feature extraction and temporal differencing into a differentiable, end-to-end trainable module. Central to our approach is a lightweight 1D convolutional encoder combined with a learnable differential temporal operator, termed Differentiable Sequence Delta (DSD), which jointly captures short-term spatial context and long-range temporal transitions. The DSD module models directional differences across sequences via a fixed-weight differencing kernel, followed by an LSTM-based refinement and optional residual projection, yielding compact, discriminative descriptors robust to viewpoint and appearance shifts. To further enhance inter-class separability, we incorporate a quadruplet loss that optimizes both positive alignment and multi-negative divergence within each batch. Unlike prior VPR methods that treat temporal aggregation as post-processing, OptiCorNet learns sequence-level embeddings directly, enabling more effective end-to-end place recognition. Comprehensive evaluations on multiple public benchmarks demonstrate that our approach outperforms state-of-the-art baselines under challenging seasonal and viewpoint variations.

cs.CV↗

arXiv2025

Perspective of high-speed Mach-Zehnder modulators based on nonlinear optics and complex band structures

Optical modulators are essential building blocks for high-capacity optical communication and massively parallel computing. Among all types of optical modulators, travelling-wave Mach-Zehnder modulators (TW-MZMs) featuring high speed and efficiency are widely used, and have been developed on a variety of integrated material platforms. Existing methods to design and simulate TW-MZMs so far strongly rely on the peculiar material properties, and thus inevitably involve complicated electrical-circuit models. As a result, these methods diverge significantly. In addition, they become increasingly inefficient and inaccurate for TW-MZMs with extending length and levitating modulation speed, posing formidable challenges for millimeter-wave and terahertz operation. Here, we present an innovative perspective to understand and analyze high-speed TW-MZMs. Our perspective leverages nonlinear optics and complex band structures of RF photonic crystals, and is thus entirely electromagnetic-wave-based. Under this perspective, we showcase the design, optoelectronic simulation and experimental validation of high-speed TW-MZMs based on Si and LiNbO$_3$, and further demonstrate unambiguous advantages in simplicity, accuracy and efficiency over conventional methods. Our approach can essentially be applied to nearly any integrated material platform, including those based on semiconductors and electro-absorption materials. With high-frequency electrode designs and optoelectronic co-simulation, our approach facilitates the synergy and convergence of electronics and photonics, and offers a viable route to constructing future high-speed millimeter-wave and terahertz photonics and quantum systems.

physics.optics↗

arXiv2025

Kwai Keye-VL Technical Report

While Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities on static images, they often fall short in comprehending dynamic, information-dense short-form videos, a dominant medium in today's digital landscape. To bridge this gap, we introduce \textbf{Kwai Keye-VL}, an 8-billion-parameter multimodal foundation model engineered for leading-edge performance in short-video understanding while maintaining robust general-purpose vision-language abilities. The development of Keye-VL rests on two core pillars: a massive, high-quality dataset exceeding 600 billion tokens with a strong emphasis on video, and an innovative training recipe. This recipe features a four-stage pre-training process for solid vision-language alignment, followed by a meticulous two-phase post-training process. The first post-training stage enhances foundational capabilities like instruction following, while the second phase focuses on stimulating advanced reasoning. In this second phase, a key innovation is our five-mode ``cold-start'' data mixture, which includes ``thinking'', ``non-thinking'', ``auto-think'', ``think with image'', and high-quality video data. This mixture teaches the model to decide when and how to reason. Subsequent reinforcement learning (RL) and alignment steps further enhance these reasoning capabilities and correct abnormal model behaviors, such as repetitive outputs. To validate our approach, we conduct extensive evaluations, showing that Keye-VL achieves state-of-the-art results on public video benchmarks and remains highly competitive on general image-based tasks (Figure 1). Furthermore, we develop and release the \textbf{KC-MMBench}, a new benchmark tailored for real-world short-video scenarios, where Keye-VL shows a significant advantage.

cs.CV↗