arXiv subjects

Liang Zhao

Publications and source records attributed to Liang Zhao.

At least 145 records · Page 8Linked to original sources

arXiv2025

Implicit Neural Compression of Point Clouds

Point clouds have gained prominence across numerous applications due to their ability to accurately represent 3D objects and scenes. However, efficiently compressing unstructured, high-precision point cloud data remains a significant challenge. In this paper, we propose NeRC$^3$, a novel point cloud compression framework that leverages implicit neural representations (INRs) to encode both geometry and attributes of dense point clouds. Our approach employs two coordinate-based neural networks: one maps spatial coordinates to voxel occupancy, while the other maps occupied voxels to their attributes, thereby implicitly representing the geometry and attributes of a voxelized point cloud. The encoder quantizes and compresses network parameters alongside auxiliary information required for reconstruction, while the decoder reconstructs the original point cloud by inputting voxel coordinates into the neural networks. Furthermore, we extend our method to dynamic point cloud compression through techniques that reduce temporal redundancy, including a 4D spatio-temporal representation termed 4D-NeRC$^3$. Experimental results validate the effectiveness of our approach: For static point clouds, NeRC$^3$ outperforms octree-based G-PCC standard and existing INR-based methods. For dynamic point clouds, 4D-NeRC$^3$ achieves superior geometry compression performance compared to the latest G-PCC and V-PCC standards, while matching state-of-the-art learning-based methods. It also demonstrates competitive performance in joint geometry and attribute compression.

cs.CV↗

arXiv2025

ToolCritic: Detecting and Correcting Tool-Use Errors in Dialogue Systems

Tool-augmented large language models (LLMs) are increasingly employed in real-world applications, but tool usage errors still hinder their reliability. We introduce ToolCritic, a diagnostic framework that evaluates and improves LLM behavior in multi-turn, tool-augmented dialogues. ToolCritic detects eight distinct error types specific to tool-calling (e.g., premature invocation, argument misalignment, and misinterpretation of tool outputs) and provides targeted feedback to the main LLM. The main LLM, assumed to have strong reasoning, task understanding and orchestration capabilities, then revises its response based on ToolCritic's feedback. We systematically define these error categories and construct a synthetic dataset to train ToolCritic. Experimental results on the Schema-Guided Dialogue (SGD) dataset demonstrate that ToolCritic improves tool-calling accuracy by up to 13% over baselines, including zero-shot prompting and self-correction techniques. This represents a promising step toward more robust LLM integration with external tools in real-world dialogue applications.

cs.AI↗

arXiv2025

DiffusionX: Efficient Edge-Cloud Collaborative Image Generation with Multi-Round Prompt Evolution

Recent advances in diffusion models have driven remarkable progress in image generation. However, the generation process remains computationally intensive, and users often need to iteratively refine prompts to achieve the desired results, further increasing latency and placing a heavy burden on cloud resources. To address this challenge, we propose DiffusionX, a cloud-edge collaborative framework for efficient multi-round, prompt-based generation. In this system, a lightweight on-device diffusion model interacts with users by rapidly producing preview images, while a high-capacity cloud model performs final refinements after the prompt is finalized. We further introduce a noise level predictor that dynamically balances the computation load, optimizing the trade-off between latency and cloud workload. Experiments show that DiffusionX reduces average generation time by 15.8% compared with Stable Diffusion v1.5, while maintaining comparable image quality. Moreover, it is only 0.9% slower than Tiny-SD with significantly improved image quality, thereby demonstrating efficiency and scalability with minimal overhead.

cs.CV↗

arXiv2025

Non-Rigid Structure-from-Motion via Differential Geometry with Recoverable Conformal Scale

Non-rigid structure-from-motion (NRSfM), a promising technique for addressing the mapping challenges in monocular visual deformable simultaneous localization and mapping (SLAM), has attracted growing attention. We introduce a novel method, called Con-NRSfM, for NRSfM under conformal deformations, encompassing isometric deformations as a subset. Our approach performs point-wise reconstruction using 2D selected image warps optimized through a graph-based framework. Unlike existing methods that rely on strict assumptions, such as locally planar surfaces or locally linear deformations, and fail to recover the conformal scale, our method eliminates these constraints and accurately computes the local conformal scale. Additionally, our framework decouples constraints on depth and conformal scale, which are inseparable in other approaches, enabling more precise depth estimation. To address the sensitivity of the formulated problem, we employ a parallel separable iterative optimization strategy. Furthermore, a self-supervised learning framework, utilizing an encoder-decoder network, is incorporated to generate dense 3D point clouds with texture. Simulation and experimental results using both synthetic and real datasets demonstrate that our method surpasses existing approaches in terms of reconstruction accuracy and robustness. The code for the proposed method will be made publicly available on the project website: https://sites.google.com/view/con-nrsfm.

cs.CV↗

arXiv2025

Cross-modal RAG: Sub-dimensional Text-to-Image Retrieval-Augmented Generation

Text-to-image generation increasingly demands access to domain-specific, fine-grained, and rapidly evolving knowledge that pretrained models cannot fully capture, necessitating the integration of retrieval methods. Existing Retrieval-Augmented Generation (RAG) methods attempt to address this by retrieving globally relevant images, but they fail when no single image contains all desired elements from a complex user query. We propose Cross-modal RAG, a novel framework that decomposes both queries and images into sub-dimensional components, enabling subquery-aware retrieval and generation. Our method introduces a hybrid retrieval strategy - combining a sub-dimensional sparse retriever with a dense retriever - to identify a Pareto-optimal set of images, each contributing complementary aspects of the query. During generation, a multimodal large language model is guided to selectively condition on relevant visual features aligned to specific subqueries, ensuring subquery-aware image synthesis. Extensive experiments on MS-COCO, Flickr30K, WikiArt, CUB, and ImageNet-LT demonstrate that Cross-modal RAG significantly outperforms existing baselines in the retrieval and further contributes to generation quality, while maintaining high efficiency.

cs.CV↗

arXiv2025

CLaw: Benchmarking Chinese Legal Knowledge in Large Language Models - A Fine-grained Corpus and Reasoning Analysis

Large Language Models (LLMs) are increasingly tasked with analyzing legal texts and citing relevant statutes, yet their reliability is often compromised by general pre-training that ingests legal texts without specialized focus, obscuring the true depth of their legal knowledge. This paper introduces CLaw, a novel benchmark specifically engineered to meticulously evaluate LLMs on Chinese legal knowledge and its application in reasoning. CLaw comprises two key components: (1) a comprehensive, fine-grained corpus of all 306 Chinese national statutes, segmented to the subparagraph level and incorporating precise historical revision timesteps for rigorous recall evaluation (64,849 entries), and (2) a challenging set of 254 case-based reasoning instances derived from China Supreme Court curated materials to assess the practical application of legal knowledge. Our empirical evaluation reveals that most contemporary LLMs significantly struggle to faithfully reproduce legal provisions. As accurate retrieval and citation of legal provisions form the basis of legal reasoning, this deficiency critically undermines the reliability of their responses. We contend that achieving trustworthy legal reasoning in LLMs requires a robust synergy of accurate knowledge retrieval--potentially enhanced through supervised fine-tuning (SFT) or retrieval-augmented generation (RAG)--and strong general reasoning capabilities. This work provides an essential benchmark and critical insights for advancing domain-specific LLM reasoning, particularly within the complex legal sphere.

cs.CL↗

arXiv2025

PL-VIWO2: A Lightweight, Fast and Robust Visual-Inertial-Wheel Odometry Using Points and Lines

Vision-based odometry has been widely adopted in autonomous driving owing to its low cost and lightweight setup; however, its performance often degrades in complex outdoor urban environments. To address these challenges, we propose PL-VIWO2, a filter-based visual-inertial-wheel odometry system that integrates an IMU, wheel encoder, and camera (supporting both monocular and stereo) for long-term robust state estimation. The main contributions are: (i) a novel line feature processing framework that exploits the geometric relationship between 2D feature points and lines, enabling fast and robust line tracking and triangulation while ensuring real-time performance; (ii) an SE(2)-constrained SE(3) wheel pre-integration method that leverages the planar motion characteristics of ground vehicles for accurate wheel updates; and (iii) an efficient motion consistency check (MCC) that filters out dynamic features by jointly using IMU and wheel measurements. Extensive experiments on Monte Carlo simulations and public autonomous driving datasets demonstrate that PL-VIWO2 outperforms state-of-the-art methods in terms of accuracy, efficiency, and robustness.

cs.RO↗

arXiv2025

StructPrune: Structured Global Pruning asymptotics with $\mathcal{O}(\sqrt{N})$ GPU Memory

Pruning is critical for scaling large language models (LLMs). Global pruning achieves strong performance but requires $\mathcal{O}(N)$ memory, which is infeasible for billion-parameter models. Local pruning reduces GPU memory usage to that of a single layer by pruning layers independently, but it neglects inter-layer dependencies and often leads to suboptimal performance in high-sparsity regimes. Unlike unstructured pruning, structured pruning produces regular sparsity patterns that align well with GPU kernels and library optimizations, making it more hardware-efficient. However, structured pruning typically relies on global pruning, since structured patterns are more prone to severe performance degradation under local optimization. To jointly achieve structured pruning and the memory efficiency of local pruning, we propose a divide-and-conquer strategy that decomposes the global pruning problem into coordinated subproblems across different modules, each of which fits within limited GPU memory. Building on this idea, we design \textbf{STRUPRUNE}, an ADMM-based framework that integrates structured sparsity into the pruning process, combining the memory efficiency of local pruning with the hardware compatibility of structured methods. We derive a closed-form analytical solution for structured pruning masks that provides an explicit rule for layer-wise sparsity allocation, and further develop an energy-based asymptotic framework yielding a softmax-form allocation scheme that simplifies optimization while adapting to heterogeneous layer importance. Experiments demonstrate that STRUPRUNE matches the perplexity of global structured pruning while reducing memory cost from $\mathcal{O}(N)$ to $\mathcal{O}(\sqrt{N})$, enabling practical deployment at the billion-parameter scale.

cs.LG↗

arXiv2025

Open Vision Reasoner: Transferring Linguistic Cognitive Behavior for Visual Reasoning

The remarkable reasoning capability of large language models (LLMs) stems from cognitive behaviors that emerge through reinforcement with verifiable rewards. This work investigates how to transfer this principle to Multimodal LLMs (MLLMs) to unlock advanced visual reasoning. We introduce a two-stage paradigm built on Qwen2.5-VL-7B: a massive linguistic cold-start fine-tuning, followed by multimodal reinforcement learning (RL) spanning nearly 1,000 steps, surpassing all previous open-source efforts in scale. This pioneering work reveals three fundamental insights: 1) Behavior transfer emerges surprisingly early in cold start due to linguistic mental imagery. 2) Cold start broadly memorizes visual behaviors, while RL critically discerns and scales up effective patterns. 3) Transfer strategically favors high-utility behaviors such as visual reflection. Our resulting model, Open-Vision-Reasoner (OVR), achieves state-of-the-art performance on a suite of reasoning benchmarks, including 95.3% on MATH500, 51.8% on MathVision and 54.6% on MathVerse. We release our model, data, and training dynamics to catalyze the development of more capable, behavior-aligned multimodal reasoners.

cs.CV↗

arXiv2025

Superfast Low Rank Approximation

Low rank approximation of a matrix (LRA) is a highly important area of Numerical Linear and Multilinear Algebra and Data Mining and Analysis. One can operate with an LRA superfast -- by using much fewer memory cells and flops than an input matrix has entries. Can we, however, compute an LRA of a matrix superfast? YES and NO. For worst case inputs, any LRA algorithm fails miserably unless it involves all input entries, but in computational practice worst case inputs seem to appear rarely, and accurate LRA are routinely computed superfast for large and important classes of matrices, in particular in the memory efficient form of CUR, widely used in data analysis. We advance formal study of this YES and NO coexistence by proving novel universal upper bounds on the spectral output error norms of all CUR LRA algorithms and, under a fixed probabilistic structure in the space of input matrices, on both spectral and Frobenius error norms of nearly all sketching LRA algorithms. These bounds imply that superfast LRA algorithms of the two kinds fail miserably only for a very narrow input class. Furthermore, in our numerical tests such superfast algorithms were consistently much more accurate than our upper estimates ensure and usually were reasonably close to optimal.

math.NA↗

arXiv2025

NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale

Prevailing autoregressive (AR) models for text-to-image generation either rely on heavy, computationally-intensive diffusion models to process continuous image tokens, or employ vector quantization (VQ) to obtain discrete tokens with quantization loss. In this paper, we push the autoregressive paradigm forward with NextStep-1, a 14B autoregressive model paired with a 157M flow matching head, training on discrete text tokens and continuous image tokens with next-token prediction objectives. NextStep-1 achieves state-of-the-art performance for autoregressive models in text-to-image generation tasks, exhibiting strong capabilities in high-fidelity image synthesis. Furthermore, our method shows strong performance in image editing, highlighting the power and versatility of our unified approach. To facilitate open research, we will release our code and models to the community.

cs.CV↗

arXiv2025

Blow-up solutions of parabolic $p$-Laplacian inequalities on locally finite graphs

In this paper, we study blow up behavior of the semilinear parabolic inequality with $p$-Laplacian operator and nonlinear source $u_t - Δ_p u \geq σ(x, t)Φ(u)$ on a locally finite connected weighted graph $G = (V, E)$. We extend the comparison principle and thereby establish the relationship between the initial value and the existence of blow-up solutions to the problem under different growth rates of $Φ$. We prove that when the growth rate of $Φ$ exceeds linear growth, blow-up solutions exist under appropriate initial conditions.

math.AP↗

arXiv2025

Step-3 is Large yet Affordable: Model-system Co-design for Cost-effective Decoding

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

cs.LG↗

arXiv2025

Low Rank Approximation at Sublinear Cost

Low Rank Approximation (LRA) of a matrix is a hot research subject, fundamental for Matrix and Tensor Computations and Big Data Mining and Analysis. Computations with low rank matrices can be performed at sublinear cost -- by using much fewer floating-point operations (flops) than an input matrix has entries, but can we compute LRA at sublinear cost? This is routinely done in computational practice for a large class of inputs, even though any sublinear cost LRA algorithm fails most miserably on worst case matrices. To provide insight into this controversy we first accelerate some popular near-optimal random sketching LRA algorithms -- to run them at sublinear cost. Then we define two probabilistic structures in the space of input matrices and estimate that the expected spectral and Frobenius error norms for the output LRA of the accelerated algorithms stay within a reasonable factor from their optima under both models, and so these sublinear cost algorithms only fail for a very narrow input class. Our upper estimates for their output accuracy are still quite high, but under some additional semi-heuristic amendments the algorithms have consistently output accurate LRA of various synthetic and real-world matrices in our numerical tests.

math.NA↗

arXiv2025

GRAG: Graph Retrieval-Augmented Generation

Naive Retrieval-Augmented Generation (RAG) focuses on individual documents during retrieval and, as a result, falls short in handling networked documents which are very popular in many applications such as citation graphs, social media, and knowledge graphs. To overcome this limitation, we introduce Graph Retrieval-Augmented Generation (GRAG), which tackles the fundamental challenges in retrieving textual subgraphs and integrating the joint textual and topological information into Large Language Models (LLMs) to enhance its generation. To enable efficient textual subgraph retrieval, we propose a novel divide-and-conquer strategy that retrieves the optimal subgraph structure in linear time. To achieve graph context-aware generation, incorporate textual graphs into LLMs through two complementary views-the text view and the graph view-enabling LLMs to more effectively comprehend and utilize the graph context. Extensive experiments on graph reasoning benchmarks demonstrate that in scenarios requiring multi-hop reasoning on textual graphs, our GRAG approach significantly outperforms current state-of-the-art RAG methods. Our datasets as well as codes of GRAG are available at https://github.com/HuieL/GRAG.

cs.LG↗

arXiv2025

CUR Low Rank Approximation of a Matrix at Sublinear Cost

Low rank approximation of a matrix (hereafter LRA) is a highly important area of Numerical Linear and Multilinear Algebra and Data Mining and Analysis. One can operate with an LRA at sublinear cost -- by using much fewer memory cells and flops than an input matrix M has entries. For worst case inputs one cannot compute even a reasonably close LRA at sublinear cost, but in computational practice accurate LRAs, even in their memory efficient form of CUR LRAs, are routinely obtained at sublinear cost for large and important classes of matrices, in particular by means of Cross-Approximation iterations, which specialize Alternating Direction techniques to LRA. We identify some classes of matrices for which CUR LRA are computed at sublinear cost as well as some sublinear cost LRA algorithms that are empirically accurate for large classes of inputs. Some of our techniques and concepts can be of independent interests.

math.NA↗

arXiv2025

Consistency-Aware Padding for Incomplete Multi-Modal Alignment Clustering Based on Self-Repellent Greedy Anchor Search

Multimodal representation is faithful and highly effective in describing real-world data samples' characteristics by describing their complementary information. However, the collected data often exhibits incomplete and misaligned characteristics due to factors such as inconsistent sensor frequencies and device malfunctions. Existing research has not effectively addressed the issue of filling missing data in scenarios where multiview data are both imbalanced and misaligned. Instead, it relies on class-level alignment of the available data. Thus, it results in some data samples not being well-matched, thereby affecting the quality of data fusion. In this paper, we propose the Consistency-Aware Padding for Incomplete Multimodal Alignment Clustering Based on Self-Repellent Greedy Anchor Search(CAPIMAC) to tackle the problem of filling imbalanced and misaligned data in multimodal datasets. Specifically, we propose a self-repellent greedy anchor search module(SRGASM), which employs a self-repellent random walk combined with a greedy algorithm to identify anchor points for re-representing incomplete and misaligned multimodal data. Subsequently, based on noise-contrastive learning, we design a consistency-aware padding module (CAPM) to effectively interpolate and align imbalanced and misaligned data, thereby improving the quality of multimodal data fusion. Experimental results demonstrate the superiority of our method over benchmark datasets. The code will be publicly released at https://github.com/Autism-mm/CAPIMAC.git.

cs.LG↗

arXiv2025

Combining Graph Neural Networks and Mixed Integer Linear Programming for Molecular Inference under the Two-Layered Model

Recently, a novel two-phase framework named mol-infer for inference of chemical compounds with prescribed abstract structures and desired property values has been proposed. The framework mol-infer is primarily based on using mixed integer linear programming (MILP) to simulate the computational process of machine learning methods and describe the necessary and sufficient conditions to ensure such a chemical graph exists. The existing approaches usually first convert the chemical compounds into handcrafted feature vectors to construct prediction functions, but because of the limit on the kinds of descriptors originated from the need for tractability in the MILP formulation, the learning performances on datasets of some properties are not good enough. A lack of good learning performance can greatly lower the quality of the inferred chemical graphs, and thus improving learning performance is of great importance. On the other hand, graph neural networks (GNN) offer a promising machine learning method to directly utilize the chemical graphs as the input, and many existing GNN-based approaches to the molecular property prediction problem have shown that they can enjoy better learning performances compared to the traditional approaches that are based on feature vectors. In this study, we develop a molecular inference framework based on mol-infer, namely mol-infer-GNN, that utilizes GNN as the learning method while keeping the great flexibility originated from the two-layered model on the abstract structure of the chemical graph to be inferred. We conducted computational experiments on the QM9 dataset to show that our proposed GNN model can obtain satisfying learning performances for some properties despite its simple structure, and can infer small chemical graphs comprising up to 20 non-hydrogen atoms within reasonable computational time.

cs.LG↗