arXiv subjects

Yuxiang Duan

Publications and source records attributed to Yuxiang Duan.

6 recordsLinked to original sources

arXiv2026

AgenticVAU: Multi-Agent Explore-Verify Reasoning for Video Anomaly Understanding

Video anomaly understanding (VAU) focuses on comprehensively interpreting abnormal events in videos, requiring models to identify anomalous occurrences, discover their supporting evidence, and explain the underlying causes beyond simple anomaly detection. Existing VAU methods often rely on specialized training or limited observations, restricting generalization or evidence coverage. Although single-agent alternatives support adaptive video observation, they still integrate exploration, observation, and decision-making within a unified reasoning process, offering limited role specialization and structured evidence coordination. To address these limitations, we present AgenticVAU, a training-free multi-agent framework that casts VAU as an explore--verify process, where the system first discovers potential anomalies and then verifies them through targeted observations. To achieve this, four specialized agents are introduced to handle visual-rule construction, search planning, video observation, and final decision, respectively. These agents communicate through an anchor registry, a shared evidence memory that binds each observation. Guided by this agent framework, AgenticVAU interleaves broad temporal exploration, dense local verification, and cross-interval comparison until sufficient evidence is collected. We conduct extensive experiments on the ECVA, UCF-Crime, and MSAD subsets of VAU-Bench, the results show that AgenticVAU outperforms zero-shot inference and reinforcement learning-based baselines, demonstrating the value of multi-agent collaboration for video anomaly understanding.

cs.CV↗

arXiv2026

M3MAD-Bench: Multi-Dimensional Evaluation of Multi-Agent Debate Across Domains and Modalities

As an agent-level reasoning and coordination paradigm, Multi-Agent Debate (MAD) orchestrates multiple agents through structured debate to improve answer quality and support complex reasoning. However, existing research on MAD suffers from two fundamental limitations: evaluations are conducted under fragmented and inconsistent settings, making fair comparison difficult, and are largely confined to text-only scenarios, leaving its effectiveness in multimodal settings underexplored. To address these gaps, we introduce M3MAD-Bench, a unified and extensible benchmark for evaluating MAD methods across Multi-domain tasks, Multi-modal inputs, and Multi-dimensional metrics. M3MAD-Bench establishes standardized protocols over five core task domains, including Knowledge, Mathematics, Medicine, Natural Sciences, and Complex Reasoning, covering a total of 13 datasets, and systematically includes both pure text and vision-language data, enabling controlled cross-modality comparison. We evaluate MAD methods on 9 base models spanning different architectures, scales, and modality capabilities. Beyond accuracy, M3MAD-Bench incorporates efficiency-oriented metrics such as token consumption and inference time, providing a holistic view of performance--cost trade-offs. Through extensive experiments, we derive nine key insights, revealing that MAD is not uniformly effective: collaborative methods are generally more robust than adversarial ones, especially on reasoning-intensive and multimodal tasks, but often incur substantial efficiency costs. These findings provide practical guidance for selecting and designing MAD strategies in real-world applications. We believe M3MAD-Bench offers a reliable foundation for future research on standardized and reproducible MAD evaluation. The code is available at https://github.com/liaolea/M3MAD-Bench.

cs.AI↗

arXiv2025

TransPrune: Token Transition Pruning for Efficient Large Vision-Language Model

Large Vision-Language Models (LVLMs) have advanced multimodal learning but face high computational costs due to the large number of visual tokens, motivating token pruning to improve inference efficiency. The key challenge lies in identifying which tokens are truly important. Most existing approaches rely on attention-based criteria to estimate token importance. However, they inherently suffer from certain limitations, such as positional bias. In this work, we explore a new perspective on token importance based on token transitions in LVLMs. We observe that the transition of token representations provides a meaningful signal of semantic information. Based on this insight, we propose TransPrune, a training-free and efficient token pruning method. Specifically, TransPrune progressively prunes tokens by assessing their importance through a combination of Token Transition Variation (TTV)-which measures changes in both the magnitude and direction of token representations-and Instruction-Guided Attention (IGA), which measures how strongly the instruction attends to image tokens via attention. Extensive experiments demonstrate that TransPrune achieves comparable multimodal performance to original LVLMs, such as LLaVA-v1.5 and LLaVA-Next, across eight benchmarks, while reducing inference TFLOPs by more than half. Moreover, TTV alone can serve as an effective criterion without relying on attention, achieving performance comparable to attention-based methods. The code will be made publicly available upon acceptance of the paper at https://github.com/liaolea/TransPrune.

cs.CV↗

arXiv2025

GridPrune: From "Where to Look" to "What to Select" in Visual Token Pruning for MLLMs

Multimodal large language models (MLLMs) have shown remarkable capabilities in a wide range of vision-language tasks. However, the large number of visual tokens introduces significant computational overhead. To address this issue, visual token pruning has emerged as a key technique for enhancing the efficiency of MLLMs. In cognitive science, humans tend to first determine which regions of a scene to attend to ("where to look") before deciding which specific elements within those regions to process in detail ("what to select"). This two-stage strategy enables the visual system to efficiently allocate attention at a coarse spatial level before performing fine-grained selection. However, existing pruning methods primarily focus on directly optimizing "what to select", typically using attention scores or similarity metrics. They rarely consider "where to look", which has been shown to lead to inefficient spatial allocation, positional bias, and the retention of irrelevant or redundant tokens. In this paper, we propose GridPrune, a method that replaces the global Top-K mechanism with a "guide-globally, select-locally" zonal selection system. GridPrune splits the pruning process into two steps: first, it uses text-conditional guidance to dynamically allocate a token budget across spatial zones; and then, it performs local selection within each budgeted zone. Experimental results demonstrate that GridPrune achieves superior performance across various MLLM architectures. On LLaVA-NeXT-7B, GridPrune retains 96.98% of the full performance while using 11.1% of the tokens, outperforming the best-performing baseline by 2.34% at the same pruning rate.

cs.CV↗

arXiv2024

Automatic Mitigation of Dynamic Atmospheric Turbulence Using Optical Phase Conjugation for Coherent Free-Space Optical Communications

Coherent detection can provide enhanced receiver sensitivity and spectral efficiency in free-space optical (FSO) communications. However, turbulence can cause modal power coupling effects on a Gaussian data beam and significantly degrade the mixing efficiency between the data beam and a Gaussian local oscillator (LO) in the coherent detector. Optical phase conjugation (OPC) in a photorefractive crystal can "automatically" mitigate turbulence by: (a) recording a back-propagated turbulence-distorted probe beam, and (b) creating a phase-conjugate beam that has the inverse phase distortion of the medium as the transmitted data beam. However, previously reported crystal-based OPC approaches for FSO links have demonstrated either: (i) a relatively fast response time of 35 ms but at a relatively low data rate (e.g., <1 Mbit/s), or (ii) a relatively high data rate of 2-Gbit/s but at a slow response time (e.g., >60 s). Here, we report an OPC approach for the automatic mitigation of dynamic turbulence that enables both a high data rate (8 Gbit/s) data beam and a rapid (<5 ms) response time. For a similar data rate, this represents a 10,000-fold faster response time than previous reports, thereby enabling mitigation for dynamic effects. In our approach, the transmitted pre-distorted phase-conjugate data beam is generated by four-wave mixing in a GaAs crystal of three input beams: a turbulence-distorted probe beam, a Gaussian reference beam regenerated from the probe beam, and a Gaussian data beam carrying a high-speed data channel. We experimentally demonstrate our approach in an 8-Gbit/s quadrature-phase-shift-keying coherent FSO link through emulated dynamic turbulence. Our results show ~10-dB improvement in the mixing efficiency of the LO with the data beam under dynamic turbulence with a bandwidth of up to ~260 Hz (Greenwood frequency).

physics.optics↗

arXiv2023

Temporally and Longitudinally Tailored Dynamic Space-Time Wave Packets

In general, space-time wave packets with correlations between transverse spatial fields and temporal frequency spectra can lead to unique spatiotemporal dynamics, thus enabling control of the instantaneous light properties. However, spatiotemporal dynamics generated in previous approaches manifest themselves at a given propagation distance yet not arbitrarily tailored longitudinally. Here, we propose and demonstrate a new versatile class of judiciously synthesized wave packets whose spatiotemporal evolution can be arbitrarily engineered to take place at various predesigned distances along the longitudinal propagation path. Spatiotemporal synthesis is achieved by introducing a 2-dimensional spectrum comprising both temporal and longitudinal wavenumbers associated with specific transverse Bessel-Gaussian fields. The resulting spectra are then employed to produce wave packets evolving in both time and axial distance - in full accord with the theoretical analysis. In this respect, various light degrees of freedom can be independently manipulated, such as intensity, polarization, and transverse spatial distribution (e.g., orbital angular momentum). Through a temporal-longitudinal frequency comb spectrum, we simulate the synthesis of the aforementioned wave packet properties, indicating a decrease in relative error compared to the desired phenomena as more spectral components are incorporated. Additionally, we experimentally demonstrate tailorable spatiotemporal fields carrying time- and longitudinal-varying orbital angular momentum, such that the local topological charge evolves every ~1 ps in the time domain and 10 cm axially. We believe that our space-time wave packets can significantly expand the exploration of spatiotemporal dynamics in the longitudinal dimension, and potentially enable novel applications in ultrafast microscopy, light-matter interactions, and nonlinear optics.

physics.optics↗