Searcharxiv⌕ Search

arXiv subjects

Lei Zhao

Publications and source records attributed to Lei Zhao.

At least 55 records · Page 3Linked to original sources
arXiv2026

TikArt: Stabilizing Aperture-Guided Fine-Grained Visual Reasoning with Reinforcement Learning

Fine-grained visual reasoning in multimodal large language models (MLLMs) is bottlenecked by single-pass global image encoding: key evidence often lies in tiny objects, cluttered regions, subtle markings, or dense charts. We present \textbf{TikArt} (\textbf{T}h\textbf{i}n\textbf{k}ing \textbf{A}pe\textbf{rt}ure), an aperture-guided agent that formulates multimodal reasoning as sequential evidence acquisition over regions of interest. TikArt follows a Think--Aperture--Observe (TAO) loop that interleaves language reasoning with two aperture actions: Zoom, which extracts rectangular crops, and Segment, which invokes an off-the-shelf segmenter to produce object-centric mask-based views for irregular targets. A mandatory Observation step after every aperture action writes local evidence back into text, yielding interpretable aperture trajectories and persistent linguistic memory. Built on Qwen3-VL-8B, TikArt is trained with GRPO-style reinforcement learning under a two-stage curriculum. To stabilize long-horizon tool-integrated learning, we introduce Relative Uncertainty Reduction (RUR), a dense reward computed by a frozen evaluator that favors evidence-building trajectories and mitigates degenerate tool use. Experiments on high-resolution reasoning, general multimodal understanding, and both referring and reasoning-oriented segmentation show consistent gains over the backbone, demonstrating that aperture-guided observation improves fine-grained visual reasoning and transfers naturally to pixel-level grounding.

cs.CV↗
arXiv2026

Structure-aware Prompt Adaptation from Seen to Unseen for Open-Vocabulary Compositional Zero-Shot Learning

The goal of Open-Vocabulary Compositional Zero-Shot Learning (OV-CZSL) is to recognize attribute-object compositions in the open-vocabulary setting, where compositions of both seen and unseen attributes and objects are evaluated. Recently, prompt tuning methods have demonstrated strong generalization capabilities in the closed setting, where only compositions of seen attributes and objects are evaluated, i.e., Compositional Zero-Shot Learning (CZSL). However, directly applying these methods to OV-CZSL may not be sufficient to generalize to unseen attributes, objects and their compositions, as it is limited to seen attributes and objects. Normally, when faced with unseen concepts, humans adopt analogies with seen concepts that have the similar semantics thereby inferring their meaning (e.g., "wet" and "damp", "shirt" and "jacket"). In this paper, we experimentally show that the distribution of semantically related attributes or objects tends to form consistent local structures in the embedding space. Based on the above structures, we propose Structure-aware Prompt Adaptation (SPA) method, which enables models to generalize from seen to unseen attributes and objects. Specifically, in the training stage, we design a Structure-aware Consistency Loss (SCL) that encourages the local structure's consistency of seen attributes and objects in each iteration. In the inference stage, we devise a Structure-guided Adaptation Strategy (SAS) that adaptively aligns the structures of unseen attributes and objects with those of trained seen attributes and objects with similar semantics. Notably, SPA is a plug-and-play method that can be seamlessly integrated into existing CZSL prompt tuning methods. Extensive experiments on OV-CZSL benchmarks demonstrate that SPA achieves competitive closed-set performance while significantly improving open-vocabulary results.

cs.CV↗
arXiv2026

An Efficient LiDAR-Camera Fusion Network for Multi-Class 3D Dynamic Object Detection and Trajectory Prediction

Service mobile robots are often required to avoid dynamic objects while performing their tasks, but they usually have only limited computational resources. To further advance the practical application of service robots in complex dynamic environments, we propose an efficient multi-modal framework for 3D object detection and trajectory prediction, which synergistically integrates LiDAR and camera inputs to achieve real-time perception of pedestrians, vehicles, and riders in 3D space.The framework incorporates two novel models: 1) a Unified modality detector with Mamba and Transformer (UniMT) for object detection, which achieves high-accuracy object detection with fast inference speed, and 2) a Reference Trajectory-based Multi-Class Transformer (RTMCT) for efficient and diverse trajectory prediction of multi-class objects with flexible-length trajectories. Evaluations on the CODa benchmark demonstrate that our method outperforms existing ones in both detection (+3.71\% in mAP) and trajectory prediction (-0.408m in minADE$_5$ of pedestrians) metrics. Furthermore, on the challenging nuScenes detection benchmark, our detection model achieves competitive performance among LiDAR-camera fusion methods, with a mAP of 72.7\% and NDS of 75.3\%. Remarkably, the system demonstrates exceptional generalizability and practical deployment potential. When transferred and implemented on a wheelchair robot with an entry-level NVIDIA RTX 3060 GPU, it achieves real-time inference at 13.9 frames per second (FPS) with satisfactory accuracy. To facilitate reproducibility and practical deployment, we release the related code of the method at \href{https://github.com/TossherO/3D_Perception}{https://github.com/TossherO/3D\_Perception} and its ROS inference version at \href{https://github.com/TossherO/ros_packages}{https://github.com/TossherO/ros\_packages}.

cs.RO↗
arXiv2026

Singpath-VL Technical Report

We present Singpath-VL, a vision-language large model, to fill the vacancy of AI assistant in cervical cytology. Recent advances in multi-modal large language models (MLLMs) have significantly propelled the field of computational pathology. However, their application in cytopathology, particularly cervical cytology, remains underexplored, primarily due to the scarcity of large-scale, high-quality annotated datasets. To bridge this gap, we first develop a novel three-stage pipeline to synthesize a million-scale image-description dataset. The pipeline leverages multiple general-purpose MLLMs as weak annotators, refines their outputs through consensus fusion and expert knowledge injection, and produces high-fidelity descriptions of cell morphology. Using this dataset, we then fine-tune the Qwen3-VL-4B model via a multi-stage strategy to create a specialized cytopathology MLLM. The resulting model, named Singpath-VL, demonstrates superior performance in fine-grained morphological perception and cell-level diagnostic classification. To advance the field, we will open-source a portion of the synthetic dataset and benchmark.

cs.CV↗
arXiv2026

MPA: Multimodal Prototype Augmentation for Few-Shot Learning

Recently, few-shot learning (FSL) has become a popular task that aims to recognize new classes from only a few labeled examples and has been widely applied in fields such as natural science, remote sensing, and medical images. However, most existing methods focus only on the visual modality and compute prototypes directly from raw support images, which lack comprehensive and rich multimodal information. To address these limitations, we propose a novel Multimodal Prototype Augmentation FSL framework called MPA, including LLM-based Multi-Variant Semantic Enhancement (LMSE), Hierarchical Multi-View Augmentation (HMA), and an Adaptive Uncertain Class Absorber (AUCA). LMSE leverages large language models to generate diverse paraphrased category descriptions, enriching the support set with additional semantic cues. HMA exploits both natural and multi-view augmentations to enhance feature diversity (e.g., changes in viewing distance, camera angles, and lighting conditions). AUCA models uncertainty by introducing uncertain classes via interpolation and Gaussian sampling, effectively absorbing uncertain samples. Extensive experiments on four single-domain and six cross-domain FSL benchmarks demonstrate that MPA achieves superior performance compared to existing state-of-the-art methods across most settings. Notably, MPA surpasses the second-best method by 12.29% and 24.56% in the single-domain and cross-domain setting, respectively, in the 5-way 1-shot setting.

cs.CV↗
arXiv2026

M$^2$-Miner: Multi-Agent Enhanced MCTS for Mobile GUI Agent Data Mining

Graphical User Interface (GUI) agent is pivotal to advancing intelligent human-computer interaction paradigms. Constructing powerful GUI agents necessitates the large-scale annotation of high-quality user-behavior trajectory data (i.e., intent-trajectory pairs) for training. However, manual annotation methods and current GUI agent data mining approaches typically face three critical challenges: high construction cost, poor data quality, and low data richness. To address these issues, we propose M$^2$-Miner, the first low-cost and automated mobile GUI agent data-mining framework based on Monte Carlo Tree Search (MCTS). For better data mining efficiency and quality, we present a collaborative multi-agent framework, comprising InferAgent, OrchestraAgent, and JudgeAgent for guidance, acceleration, and evaluation. To further enhance the efficiency of mining and enrich intent diversity, we design an intent recycling strategy to extract extra valuable interaction trajectories. Additionally, a progressive model-in-the-loop training strategy is introduced to improve the success rate of data mining. Extensive experiments have demonstrated that the GUI agent fine-tuned using our mined data achieves state-of-the-art performance on several commonly used mobile GUI benchmarks. Our work will be released to facilitate the community research.

cs.AI↗
arXiv2026

Noise-immune and AI-enhanced DNA storage via adaptive partition mapping of digital data

Encoding digital information into DNA sequences offers an attractive potential solution for storing rapidly growing data under the information age and the rise of artificial intelligence. However, practical implementations of DNA storage are constrained by errors introduced during synthesis, preservation, and sequencing processes, and traditional error-correcting codes remain vulnerable to noise levels that exceed predefined thresholds. Here, we developed a Partitioning-mapping with Jump-rotating (PJ) encoding scheme, which exhibits exceptional noise resilience. PJ removes cross-strand information dependencies so that strand loss manifests as localized gaps rather than catastrophic file failure. It prioritizes file decodability under arbitrary noise conditions and leverages AI-based inference to enable controllable recovery of digital information. For the intra-strand encoding, we develop a jump-rotating strategy that relaxes sequence constraints relative to conventional rotating codes and provides tunable information density via an adjustable jump length. Based on this encoding architecture, the original file information can always be decoded and recovered under any strand loss ratio, with fidelity degrading smoothly as damage increases. We demonstrate that original files can be effectively recovered even with 10% strand loss, and machine learning datasets stored under these conditions retain their classification performance. Experiments further confirmed that PJ successfully decodes image files after extreme environmental disturbance using accelerated aging and high-intensity X-ray irradiation. By eliminating reliance on prior error probabilities, PJ establishes a general framework for robust, archival DNA storage capable of withstanding the rigorous conditions of real-world preservation.

cs.IT↗
arXiv2026

4D-ARE: Bridging the Attribution Gap in LLM Agent Requirements Engineering

We deployed an LLM agent with ReAct reasoning and full data access. It executed flawlessly, yet when asked "Why is completion rate 80%?", it returned metrics instead of causal explanation. The agent knew how to reason but we had not specified what to reason about. This reflects a gap: runtime reasoning frameworks (ReAct, Chain-of-Thought) have transformed LLM agents, but design-time specification--determining what domain knowledge agents need--remains under-explored. We propose 4D-ARE (4-Dimensional Attribution-Driven Agent Requirements Engineering), a preliminary methodology for specifying attribution-driven agents. The core insight: decision-makers seek attribution, not answers. Attribution concerns organize into four dimensions (Results -> Process -> Support -> Long-term), motivated by Pearl's causal hierarchy. The framework operationalizes through five layers producing artifacts that compile directly to system prompts. We demonstrate the methodology through an industrial pilot deployment in financial services. 4D-ARE addresses what agents should reason about, complementing runtime frameworks that address how. We hypothesize systematic specification amplifies the power of these foundational advances. This paper presents a methodological proposal with preliminary industrial validation; rigorous empirical evaluation is planned for future work.

cs.SE↗
arXiv2025

VLA-RAIL: A Real-Time Asynchronous Inference Linker for VLA Models and Robots

Vision-Language-Action (VLA) models have achieved remarkable breakthroughs in robotics, with the action chunk playing a dominant role in these advances. Given the real-time and continuous nature of robotic motion control, the strategies for fusing a queue of successive action chunks have a profound impact on the overall performance of VLA models. Existing methods suffer from jitter, stalling, or even pauses in robotic action execution, which not only limits the achievable execution speed but also reduces the overall success rate of task completion. This paper introduces VLA-RAIL (A Real-Time Asynchronous Inference Linker), a novel framework designed to address these issues by conducting model inference and robot motion control asynchronously and guaranteeing smooth, continuous, and high-speed action execution. The core contributions of the paper are two fold: a Trajectory Smoother that effectively filters out the noise and jitter in the trajectory of one action chunk using polynomial fitting and a Chunk Fuser that seamlessly align the current executing trajectory and the newly arrived chunk, ensuring position, velocity, and acceleration continuity between two successive action chunks. We validate the effectiveness of VLA-RAIL on a benchmark of dynamic simulation tasks and several real-world manipulation tasks. Experimental results demonstrate that VLA-RAIL significantly reduces motion jitter, enhances execution speed, and improves task success rates, which will become a key infrastructure for the large-scale deployment of VLA models.

cs.RO↗
arXiv2025

RLLaVA: An RL-central Framework for Language and Vision Assistants

We present an RL-central framework for Language and Vision Assistants (RLLaVA) with its formulation of Markov decision process (MDP). RLLaVA decouples RL algorithmic logic from model architecture and distributed execution, supporting researchers in implementing new RL algorithms with minimal code, and to plug in a broad family of RL methods and vision-language models (VLMs) while remaining agnostic to specific training and inference engines. RLLaVA makes resource-efficient training of 1B--7B models feasible on common GPUs; notably, 4B-scale models can be trained end-to-end with full-parameter updates on a single 24GB GPU. Experiments on multi-modal and agentic tasks demonstrate that RLLaVA has task extensibility, and the models trained with it consistently improve performance over base models, competitive with other specially engineered RL frameworks. The code is available at https://github.com/TinyLoopX/RLLaVA.

cs.LG↗
arXiv2025

Simulation of MAPS and a MAPS-based Inner Tracker for the Super Tau-Charm Facility

Monolithic Active Pixel Sensors (MAPS) are a promising detector candidate for the inner tracker of the Super Tau-Charm Facility (STCF). To evaluate the performance of MAPS and the MAPS-based inner tracker, a dedicated simulation workflow has been developed, offering essential insights for detector design and optimization. The intrinsic characteristics of MAPS, designed using several fabrication processes and pixel geometries, were investigated through a combination of Technology Computer Aided Design (TCAD) and Monte Carlo simulations. Simulations were conducted with both minimum ionizing particles and $^{55}$Fe X-rays to assess critical parameters such as detection efficiency, cluster size, spatial resolution, and charge collection efficiency. Based on these evaluations, a MAPS sensor featuring a strip-like pixel and a high-resistivity epitaxial layer is selected as the baseline sensor design for the STCF inner tracker due to its excellent performance. Using this optimized MAPS design, a three-layer MAPS-based inner tracker was modeled and simulated. The simulation demonstrated an average detection efficiency exceeding 99%, spatial resolutions of 44.8$\rm{μm}$ in the $z$ direction and 8.2$\rm{μm}$ in the $r-ϕ$ direction, and an intrinsic sensor time resolution of 5.9ns for 1GeV/c $μ^-$ particles originating from the interaction point. These promising results suggest that the MAPS-based inner tracker fulfills the performance requirements of the STCF experiment.

physics.ins-det↗
arXiv2025

A common first integral from three-body secular theory and Kepler billiards

We observe that a particular first integral of the partially-averaged system in the secular theory of the three-body problem appears also as an important conserved quantity of integrable Kepler billiards. In this note we illustrate their common roots with the projective dynamics of the two-center problem. We then combine these two aspects to define a class of integrable billiard systems on surfaces of constant curvature.

math.DS↗
arXiv2025

Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation

Sequential recommendation has garnered significant attention for its ability to capture dynamic preferences by mining users' historical interaction data. Given that users' complex and intertwined periodic preferences are difficult to disentangle in the time domain, recent research is exploring frequency domain analysis to identify these hidden patterns. However, current frequency-domain-based methods suffer from two key limitations: (i) They primarily employ static filters with fixed characteristics, overlooking the personalized nature of behavioral patterns; (ii) While the global discrete Fourier transform excels at modeling long-range dependencies, it can blur non-stationary signals and short-term fluctuations. To overcome these limitations, we propose a novel method called Wavelet Enhanced Adaptive Frequency Filter for Sequential Recommendation. Specifically, it consists of two vital modules: dynamic frequency-domain filtering and wavelet feature enhancement. The former is used to dynamically adjust filtering operations based on behavioral sequences to extract personalized global information, and the latter integrates wavelet transform to reconstruct sequences, enhancing blurred non-stationary signals and short-term fluctuations. Finally, these two modules work to achieve comprehensive performance and efficiency optimization in long sequential recommendation scenarios. Extensive experiments on four widely-used benchmark datasets demonstrate the superiority of our work.

cs.IR↗
arXiv2025

MAPo : Motion-Aware Partitioning of Deformable 3D Gaussian Splatting for High-Fidelity Dynamic Scene Reconstruction

3D Gaussian Splatting, known for enabling high-quality static scene reconstruction with fast rendering, is increasingly being applied to multi-view dynamic scene reconstruction. A common strategy involves learning a deformation field to model the temporal changes of a canonical set of 3D Gaussians. However, these deformation-based methods often produce blurred renderings and lose fine motion details in highly dynamic regions due to the inherent limitations of a single, unified model in representing diverse motion patterns. To address these challenges, we introduce Motion-Aware Partitioning of Deformable 3D Gaussian Splatting (MAPo), a novel framework for high-fidelity dynamic scene reconstruction. Its core is a dynamic score-based partitioning strategy that distinguishes between high- and low-dynamic 3D Gaussians. For high-dynamic 3D Gaussians, we recursively partition them temporally and duplicate their deformation networks for each new temporal segment, enabling specialized modeling to capture intricate motion details. Concurrently, low-dynamic 3DGs are treated as static to reduce computational costs. However, this temporal partitioning strategy for high-dynamic 3DGs can introduce visual discontinuities across frames at the partition boundaries. To address this, we introduce a cross-frame consistency loss, which not only ensures visual continuity but also further enhances rendering quality. Extensive experiments demonstrate that MAPo achieves superior rendering quality compared to baselines while maintaining comparable computational costs, particularly in regions with complex or rapid motions.

cs.CV↗
arXiv2025

Forgetting by Pruning: Data Deletion in Join Cardinality Estimation

Machine unlearning in learned cardinality estimation (CE) systems presents unique challenges due to the complex distributional dependencies in multi-table relational data. Specifically, data deletion, a core component of machine unlearning, faces three critical challenges in learned CE models: attribute-level sensitivity, inter-table propagation and domain disappearance leading to severe overestimation in multi-way joins. We propose Cardinality Estimation Pruning (CEP), the first unlearning framework specifically designed for multi-table learned CE systems. CEP introduces Distribution Sensitivity Pruning, which constructs semi-join deletion results and computes sensitivity scores to guide parameter pruning, and Domain Pruning, which removes support for value domains entirely eliminated by deletion. We evaluate CEP on state-of-the-art architectures NeuroCard and FACE across IMDB and TPC-H datasets. Results demonstrate CEP consistently achieves the lowest Q-error in multi-table scenarios, particularly under high deletion ratios, often outperforming full retraining. Furthermore, CEP significantly reduces convergence iterations, incurring negligible computational overhead of 0.3%-2.5% of fine-tuning time.

cs.DB↗
arXiv2025

GContextFormer: A global context-aware hybrid multi-head attention approach with scaled additive aggregation for multimodal trajectory prediction

Multimodal trajectory prediction generates multiple plausible future trajectories to address vehicle motion uncertainty from intention ambiguity and execution variability. However, HD map-dependent models suffer from costly data acquisition, delayed updates, and vulnerability to corrupted inputs, causing prediction failures. Map-free approaches lack global context, with pairwise attention over-amplifying straight patterns while suppressing transitional patterns, resulting in motion-intention misalignment. This paper proposes GContextFormer, a plug-and-play encoder-decoder architecture with global context-aware hybrid attention and scaled additive aggregation achieving intention-aligned multimodal prediction without map reliance. The Motion-Aware Encoder builds scene-level intention prior via bounded scaled additive aggregation over mode-embedded trajectory tokens and refines per-mode representations under shared global context, mitigating inter-mode suppression and promoting intention alignment. The Hierarchical Interaction Decoder decomposes social reasoning into dual-pathway cross-attention: a standard pathway ensures uniform geometric coverage over agent-mode pairs while a neighbor-context-enhanced pathway emphasizes salient interactions, with gating module mediating their contributions to maintain coverage-focus balance. Experiments on eight highway-ramp scenarios from TOD-VT dataset show GContextFormer outperforms state-of-the-art baselines. Compared to existing transformer models, GContextFormer achieves greater robustness and concentrated improvements in high-curvature and transition zones via spatial distributions. Interpretability is achieved through motion mode distinctions and neighbor context modulation exposing reasoning attribution. The modular architecture supports extensibility toward cross-domain multimodal reasoning tasks. Source: https://fenghy-chen.github.io/sources/.

cs.AI↗
arXiv2025

Projective deduction of the non-trivial first integral to the Euler problem: an explicit computation

The validity of Kepler Laws for the {\it spherical Kepler problem} -- namely, the problem of the motion of a particle on the unit sphere {in $\mathbb R^3$} undergoing an attraction by another particle in the sphere, tangent to the geodesic line between the two and inversely proportional to its squared length -- prompted geometers to try to interpret such system as a '' projection'' of the familiar Kepler problem in the plane, with the hosting plane given by some affine plane in $\mathbb R^3$. At this respect, the most convenient mutual sphere-plane position has been object of a long debate, an account of which can be found in \cite{Albouy2013}. This fascinating topic, resumed %subject, firstly by A. Albouy in the aforementioned paper, has been expanded from the theoretical side in \cite{Albouy2015}. Further investigations recently appeared in \cite{AlbouyZhao2019, Zhao1, TakeuchiZhao1, TakeuchiZhao2}. As remarked in \cite{Albouy2013, Albouy2015}, extensions of the procedure to more dynamical systems would open to the possibility of finding first integrals to a given dynamical system simply looking at the energy of the mirror problem. In this note, we focus on the case of the problem of two fixed centers, already mentioned in \cite{Albouy2013}. We provide a{n explicit} geometrical construction allowing to interpret the first integral of the problem as the energy of its projection on an ellipsoid. Compared to previous papers on the same subject, ours -- besides being based on a somehow different construction -- includes complete explicit computations. {A byproduct of our construction is the existence of two integrable mirror problems (equivalently, three quadratic integrals, including the energy) for the Kepler problem, which is an aspect of its super-integrability.

math-ph↗
arXiv2025

NL-DPE: An Analog In-memory Non-Linear Dot Product Engine for Efficient CNN and LLM Inference

Resistive Random Access Memory (RRAM) based in-memory computing (IMC) accelerators offer significant performance and energy advantages for deep neural networks (DNNs), but face three major limitations: (1) they support only \textit{static} dot-product operations and cannot accelerate arbitrary non-linear functions or data-dependent multiplications essential to modern LLMs; (2) they demand large, power-hungry analog-to-digital converter (ADC) circuits; and (3) mapping model weights to device conductance introduces errors from cell nonidealities. These challenges hinder scalable and accurate IMC acceleration as models grow. We propose NL-DPE, a Non-Linear Dot Product Engine that overcomes these barriers. NL-DPE augments crosspoint arrays with RRAM-based Analog Content Addressable Memory (ACAM) to execute arbitrary non-linear functions and data-dependent matrix multiplications in the analog domain by transforming them into decision trees, fully eliminating ADCs. To address device noise, NL-DPE uses software-based Noise Aware Fine-tuning (NAF), requiring no in-device calibration. Experiments show that NL-DPE delivers 28X energy efficiency and 249X speedup over a GPU baseline, and 22X energy efficiency and 245X speedup over existing IMC accelerators, while maintaining high accuracy.

cs.AR↗