TodayNewsResearchFieldsStacksMarketAnalytics
⌘K

Research

New papers ranked by attention, with their code, models and discussion linked. Pick a field on the left.

WeekMonthQuarterAll
NewestMost relevant

Papers

Open problems

Rising research topics

Ask the research

Sample
ByteTrail · AI and developer-tools intelligence
NewsResearchStacksAPI
Fields · this week
All fields1915Multimodal406AI agents360Frontier LLMs240Inference & serving169Teacher policy distillation134Medical image segmentation111Physics informed neural103Supervised & diffusion models71Voice AI44Forecasting time37Machine learned interatomic32Remote sensing31Electrocardiogram ecg26Federated learning23Deepfake detection23All 25 fields

Open problems across fields

  • Interface lock-in in multimodal spatial reasoning1 paper raises it · 1 addresses it

    Specializing a model on one spatial task (e.g., side detection) degrades performance on others (e.g., distance estimation), limiting general spatial intelligence.

    • ChronoReplay: One Answer Interface Is Not Enough for Streaming Spatial Intelligenceraises · addresses
  • Lack of benchmarks for agent harness reliability1 paper raises it · 1 addresses it

    Existing evaluations focus on model capabilities, not the reliability of the systems that deploy and manage agents.

    • Trial by Trace: Benchmarking AI Agent Harnesses for Reliability and Failure Attributionraises · addresses
  • Negative net value of agent memory systems1 paper raises it · 1 addresses it

    Memory systems often underperform naive retrieval, as their design can obscure or remove critical evidence from raw history.

    • Memory Tax: Dissecting the Net Value of Agent Memoryraises · addresses
  • Credit assignment in diffusion language models1 paper raises it · 2 address it

    Iterative denoising in DLMs makes it hard to attribute rewards to individual tokens, limiting reinforcement learning effectiveness.

    • DTCA: Denoising Token Credit Assignment for Reinforcement Learning in Diffusion Language Modelsraises · addresses
    • Multiscale Policy Optimization for Long-Horizon LLM Agentsaddresses
  • Uncertainty quantification failure under class imbalance1 paper raises it · 1 addresses it

    Standard uncertainty measures fail to reflect clinical ambiguity in imbalanced datasets, systematically penalizing minority classes.

    • When uncertainty quantification fails: structural vulnerabilities and corrections for selective prediction under clinical class imbalanceraises · addresses

Named by an AI from these papers' abstracts.

Rising research topics

  • Channel state information13+44.4%
  • AI tutors13+30.0%
  • Multimodal emotion70.0%
  • Anomaly detectors18−10.0%
  • Character recognition OCR9−10.0%
  • Physics informed neural103−29.9%
  • Machine learned interatomic32−31.9%
  • Medical image segmentation111−38.0%

Papers

  • arXiv cs.CVOct 11▲ 33

    Data-Centric Recursive Self-Improvement for Navigation via Policy-Synchronized Language-Action Dual Bootstrapping

    Songze Li, Haoming Gu, Ziyang Gong, Zhitong Wang, Gengze Zhou +9

    Vision-language navigation requires agents to follow natural-language instructions to reach target locations, making high-quality language supervision essential. Since human annotation is costly, prior work has explored scalable data synthesis methods. However, existing methods typically either use separate models for navigation and instruction generation or couple the two tasks through task-specific components; most also annotate the full trajectory corpus with a fixed model before updating, which can cause generated supervision to lag behind the evolving policy. Therefore, we introduce RSI-Nav, a data-centric recursive self-improvement framework for navigation via policy-synchronized language-action dual bootstrapping, which unifies vision-language-to-action instruction following and vision-action-to-language instruction generation within a shared model. Specifically, after supervised fine-tuning establishes fundamental capabilities, RSI-Nav progressively processes unlabeled trajectories in small shards. The current model generates instructions for these trajectories, executes them to evaluate goal completion and path efficiency, retains qualified samples, and immediately…

    CodePDF
  • arXiv cs.LGOct 11

    Heterogeneous Structured Pruning Under Resource Constraints for Hardware-Efficient Large Language Models

    Cameron Barker, Andreas Paraskeva, Jan N. van Rijn, Elliot J. Crowley, Henry Gouk

    Training-free pruning removes weights from pretrained large language models to reduce their deployment cost. However, existing approaches typically consider only the total number of retained weights. Consequently, their pruning decisions do not account for how target hardware stores and processes those weights. These methods select a sparsity structure in advance and apply it uniformly to all projections, rather than accounting for how each structure trades pruning error against latency and memory across sublayers. We propose Heterogeneous Structured Pruning (HSP), a hardware-aware pruning framework that combines two components: (i) a heterogeneous search space of dense, unstructured, 2:4, block-sparse, and head / channel (H/C) pruning options and (ii) structure-regularising constraints based on latency and memory models. Our contribution generalises existing importance-based methods: a global solver selects one option per sublayer to minimise proxy pruning loss under a resource budget, thereby assigning each projection a structure and density. We evaluate HSP on OPT and LLaMA models for an NVIDIA Jetson Orin Nano Super edge device, where it improves the quality-resource…

    PDF
  • arXiv cs.LGOct 11

    Who Sets the Pace of Judge Hacking? Judge Capability versus the Policy-Judge Gap

    Haosen Wang

    Under GRPO against a frozen, reference-free LLM judge, RL exploits the judge's false positives. Does the policy-judge capability gap or the judge itself set how fast and how far? We train 0.5B and 1.5B Qwen2.5 policies on GSM8K against four judges at least as large, and the 1.5B policy also against one smaller judge. In a post-hoc analysis, judge acceptance rose, but no cell of 12 gained more than 0.04 in seed-mean held-out greedy accuracy. A gold-reward control with the same recipe raised the 1.5B policy's accuracy by +0.054 (for the 0.5B policy, gold reward gave no detectable gain either). The depth of judge hacking tracks the judge: the plateau correlates with judge AUC (Spearman -0.69 over 12 cells), and within a family the larger judge is fooled less deeply. For the pace, two pre-specified pairs contradict the gap account: the cell with the larger gap was fooled 3.5x sooner at the pre-specified threshold. With 12 cells, however, we cannot formally separate judge AUC from the gap for either endpoint. A blind test of step-0 predictors on three held-out judges was inconclusive. At 7-8B, judges from three families plateaued between 0.37 and 0.85.

    PDF
  • arXiv cs.CLOct 11▲ 7

    RSIPO: Recursive Self-Improving Policy Optimization for Agentic Reinforcement Learning via Skill-Policy Co-Evolution

    Fan Zhang, Hao Wu, Jinyang Wu, Hongyu Qu, Zheng Lian +5

    Recent focus in agentic reinforcement learning (RL) seeks to improve credit assignment in long-horizon tasks. One promising direction combines reinforcement learning with verifiable rewards (RLVR) and on-policy self-distillation (OPSD), integrating trajectory-level advantages with dense token-level supervision. This line of work often conditions a teacher model on privileged context, such as agentic skills, to score tokens generated by the student policy. Despite their progress, they largely overlook the on-policy evolution of agentic skills and the reliability of teacher supervision. Towards this end, we introduce Recursive Self-Improving Policy Optimization (RSIPO), a self-evolving agentic RL framework that enables agentic skills and the policy model to co-evolve throughout the training process. Specifically, RSIPO enables the policy itself to extract skills from completed trajectories and explicitly optimizes its skill-generation capability through trajectory replay. It further employs empirical Bayes shrinkage to dynamically calibrate the student's trust in teacher scores, accounting for the fact that skill-enhanced teacher guidance can sometimes be less reliable. Experiments…

    Code
  • arXiv cs.ROOct 11

    Hand2Robot: Object-Centric 4D Hand Reconstruction from Human Demonstrations for Robotics

    Zetao Du, Weiqing Wang, Hanbing Zhang, Yimiao Zhou, Ye Shi +2

    Human demonstrations can capture task-relevant manipulation without using the target robot, but the recorded hand motions cannot directly serve as robot action supervision. Hand reconstruction and motion retargeting provide a direct route to such supervision, yet recovering hand motion alone does not ensure physically consistent hand-object interactions. To address this issue, we introduce Hand2Robot, an object-centric 4D hand reconstruction framework for robot policy post-training. Given videos of humans performing the target task, Hand2Robot extracts interaction clips, reconstructs object geometry, and generates robot action supervision through hand-object interaction optimization and motion retargeting. For object tracking, we combine observation confidence with temporal motion constraints to select a globally consistent object pose trajectory from per-frame pose candidates. For hand refinement, we use object geometry and motion as a reference and combine staged 2D alignment with object-local contact constraints to refine hand trajectories and finger joint configurations. Experiments on DexYCB and HO3D-v3 show improved object pose accuracy and hand-object interaction quality.…

    PDF
  • arXiv cs.ROOct 11

    SurgiCLAW: An Interpretable Atomic-Skill Agent for Precision Robotic Micromanipulation

    Angen Ye, junjie xie, Yujia Song, Junkai Huang, Dapeng Zhang

    Recent multimodal large language models (MLLMs) and vision--language--action models have provided robots with new capabilities for long-horizon task understanding and planning. In surgical robotics, however, end-to-end prediction of low-level actions remains limited by insufficient precision, opaque decisions, and scarce data. This paper presents SurgiCLAW, a hierarchical skill-based framework for precision surgical manipulation. SurgiCLAW confines an MLLM to intraoperative visual understanding, structured explanation, and atomic surgical-skill selection, while delegating physical execution to verifiable deterministic control programs. For brain--machine-interface implantation, a representative high-precision surgical task, we construct a state-to-skill invocation dataset and expand it through visual and state perturbations. The model first learns structured skill invocation through supervised fine-tuning and is then optimized using three rewards for action correctness, operation prerequisites, and format validity, improving action selection, operational safety, and output parseability. We evaluate subtask success, offline model comparison, real-robot performance, and ablations.…

    PDF
  • arXiv cs.CVOct 11

    SlotAVS: Object-Centric Audio-Visual Segmentation via Bi-Modal Slot Attention

    Arnav Sharma, Anand Mishra, Renu Rameshan

    Audio-Visual Segmentation (AVS) requires predicting pixel-wise masks of sounding objects by grounding acoustic signals in visual space. In multi-source scenes, concurrent sound sources often remain entangled within shared visual representations, leading to merged masks or unstable specialization when using standard alignment-focused transformer architectures. To address this, we present SlotAVS, an object-centric framework driven by competitive bi-modal slot attention. While robust foundation models provide rich semantic priors, SlotAVS shifts the paradigm from global cross-modal fusion to explicit competitive binding. By initializing discrete object slots from temporal audio sequences and enforcing strict slot-wise competition over spatial visual tokens, our method structurally mitigates the blending of distinct acoustic entities. Further, to stabilize instance-level learning, we introduce a Hungarian matching strategy that optimally aligns slots with ground-truth objects using spatial and boundary overlap. Coupled with a numerically stable, confidence-aware mask fusion mechanism, SlotAVS robustly aggregates multi-source predictions. Extensive experiments on the AVSBench dataset…

    PDF
  • arXiv cs.SEOct 11▲ 36

    A Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software Evolution

    Alan Wu, Wei Tao, Yiwen Guo

    Large language model agents are emerging as an important new class of users of desktop applications. However, existing desktop interfaces were designed for human end users or application developers, leaving agents to interpret visual controls or discover and compose low-level software operations. This mismatch can hinder even capable agents, yet it remains unclear what interface properties make desktop software easier for agents to use. To study this question, we treat software evolution as a natural source of interface variation and introduce DeskIF-Bench, a benchmark of 255 artifact-verified tasks across six desktop applications. By holding the task, agent, and verifier fixed while varying the software release, we measure how interface changes affect agent performance. Across two models and six applications, the newest release consistently outperforms the oldest. Trajectory analysis identifies three relevant interface properties: discoverability, granularity, and observability. Guided by these properties, we develop ANIS, a framework that synthesizes agent-native interfaces from the capabilities of installed applications. On an expanded evaluation of 368 tasks across eight…

    Code
  • arXiv cs.CLOct 11

    Robust Marathi Sentiment Analysis using Domain-Adaptive Pre-training and Data Augmentation Approaches

    Tanvi Somani, Aryan Babare, Samyak Bora, Raviraj Joshi, Abhinay Dhamankar

    For languages with limited resources, Sentiment Analysis remains a challenging problem due to limited annotated data and strong domain variability across textual sources. Marathi, despite being widely spoken, lacks extensive multi-domain sentiment analysis studies. Recent efforts by L3Cube have introduced multi-domain datasets in Marathi, enabling a better study of domain-aware sentiment classification. This study uses the pretrained MahaBERT model to improve sentiment classification for Marathi across different domains. Models are trained on individual domain datasets and domain-adaptive pretraining is applied using masked language modeling to help the model better understand domain-specific language. In addition, back-translation is used as a method of data augmentation to increase the diversity of training data. Results show that these methods lead to consistent improvements in accuracy across all domains. The improvements range from a minimum of about 0.12% to a maximum of about 1.74%, depending on the dataset. These results demonstrate that domain-adaptive pretraining and data augmentation are effective strategies for improving sentiment analysis models. This helps achieve…

    PDF
  • arXiv cs.LGOct 11

    Hyperbolic Time Series Forecasting with Learnable Decomposition and Manifold Dynamics

    Sarang Rajendra Patil, Vishal Fenn, Mengjia Xu

    Multiscale structure is central to time series forecasting: temporal signals contain slowly varying trends, intermediate seasonal patterns, and high-frequency residuals. Traditional forecasters model this multiscale structure in Euclidean space, without representing the ordering it induces among components. We identify that cascaded decomposition orders components by their characteristic frequencies, forming a natural frequency hierarchy. This hierarchy aligns with the exponential volume growth of hyperbolic space, which provides increasing representational capacity with radius and naturally accommodates nested hierarchies. We propose HyperForecast, the first long-term time series forecasting (LTSF) framework operating in hyperbolic space. HyperForecast embeds decomposed temporal segments on the Poincaré ball where radial distance captures this nested structure, from low-frequency components near the origin to high-frequency components near the boundary. We model geodesic velocity through two manifold-dynamics variants: (a) autoregressive forecasting via Poincaré residual dynamics, and (b) non-autoregressive forecasting that predicts all future segments in parallel. HyperForecast…

    PDF
  • arXiv cs.CLOct 11

    Enhancing Zero-Shot Domain Transfer in Low-Resource Sentiment Analysis: A Case Study in Marathi

    Tanvi Somani, Aryan Babare, Samyak Bora, Raviraj Joshi

    Sentiment analysis is a key task in Natural Language Processing that determines the polarity of opinions in text. For low-resource languages like Marathi--where limited data prevents the direct use of standard statistical methods--models are often trained on single-domain datasets. While this yields strong in-domain performance, it leads to poor generalization to unseen domains. Recent studies highlight the need for domain-robust systems that can handle diverse real-world content through multi-domain modeling. This study investigates how multi-domain training improves sentiment classification for unseen Marathi domains. Models are trained using single-domain, two-domain, and three-domain dataset combinations. Results show that two-domain training improves accuracy on unseen domains by about 8\% compared to single-domain training. Adding a third domain provides further improvement, though the gain is smaller, averaging around 2\% for some domains. These results suggest that training with data from multiple domains helps build more robust sentiment analysis models for low-resource languages, though the benefits reduce as more domains are added.

    PDF
  • arXiv cs.CVOct 11

    Frozen History: When Temporal Losses Stop Accumulation in Learned Real-Time Supersampling

    Hao Yu, Guoying Zhao

    Learned real-time supersampling accumulates jittered low-resolution samples along motion vectors. Many methods use a temporal consistency loss to reduce flicker. We show that with accurate registration, this loss can suppress accumulation: on static content, the network copies its warped history at almost no temporal cost. Camera jitter in motion vectors raises this cost and can hide the effect. We identify four registration errors that training can conceal: jitter-carrying motion vectors, one motion field for both resolution lattices, jittered depth, and an unregistered skip path. A model-free oracle shows that correct registration makes freezing two orders of magnitude cheaper on static pixels. Controlled training shows that registration reverses the temporal weight's effect on PSNR. Under exact registration, increasing the weight reduces pixel variance at the expense of accuracy and perceived quality, with a 0.7 to 0.9 dB loss on static frames at weight 0.25. This reversal recurs across three network families and two scales; the loss also reduces accuracy on unjittered STSS data. With a small temporal weight and a normalised sample accumulator, a plain registered network…

    PDF
  • arXiv cs.AIOct 11

    Google Earth AI Geospatial Reasoning Agent: Capabilities and a Systematic Evaluation

    Noam Slonim, Aviv Slobodkin, Gil Fidel, Gal Moshitch, Mimi Sun +32

    Google Earth AI is a family of geospatial AI models, together with an agent that orchestrates them. We present this agent, Google Earth AI's Geospatial Reasoning agent (GRA), a Gemini-powered research system that combines geospatial data sources, foundation models, and tools to answer questions about our planet. GRA organizes these into modular domain skills and pairs them with upfront geospatial planning to decompose complex queries into multi-stage workflows. Evaluating such systems is hard: real-world problems admit multiple valid workflows, require data discovery across large catalogs, and rely on evolving ground truth. We introduce two benchmarks and release their development splits. GeoExpertQueries poses complex analytical problems grounded in practitioner use cases and refined by geospatial experts; each is paired with a consensus golden workflow from three independent experts and scored by a human-validated, tool-agnostic autorater. GeoKTools consists of templated queries with deterministic reference solvers that systematically vary the number of required tool types. We also use two external benchmarks, CBGB and UnivEARTH; the latter we find to be saturated. To isolate…

    PDF
  • arXiv cs.CVOct 11

    SpikeOcc: Towards Resilient Multimodal 3D Occupancy Perception via Spiking Experts

    Yiding Sun, Zijie Xu, Hongqiang Lin, Jiajun Cao, Dongxu Zhang +5

    Multimodal 3D occupancy prediction underpins safe autonomous driving through comprehensive scene understanding. However, existing methods have largely pursued higher accuracy through sophisticated fusion, leaving resilience to missing sensors underexplored. In this paper, we investigate spiking neural networks (SNNs) as a means of building more resilient multimodal perception. Still, the discrete nature of spiking neurons makes optimization challenging. To overcome this difficulty, SpikeOcc is introduced with two complementary components, a Spiking Mixture-of-Experts (SMoE) and Sensor Availability Learning (SAL). Specifically, SMoE organizes RGB, LiDAR, and joint fusion into specialized pathways, enabling full-width unimodal representations while retaining the original multimodal features via a bounded residual. Beyond sensor dropout, SAL employs a staged curriculum with explicit routing supervision, progressively improving robustness under severe sensing conditions. Remarkably, SAL also yields notable gains for conventional models, highlighting its broader potential for resilient multimodal learning. Extensive experiments demonstrate that SpikeOcc maintains competitive…

    PDF
  • arXiv cs.CVOct 11

    RSAJEV: Training-Free Abstaining Prototype Memory for Eliminating Silent Hallucination in JEV-Based Remote Sensing Classification

    Man Chen, Antai Chen, Shuai Xu, Chengjie Gu, Huaping Zhou +3

    Scene classification of remote sensing imagery with multimodal large language models (MLLMs) is formulated as a single-forward-pass decision problem: candidate class names are lettered options in the prompt, a registered answer slot is conditioned on the prompt, and classification is read out by scoring letter tokens with the language-modeling head. We show that although this letter binding is highly effective within its training vocabulary (97.71% OA on UCM), it behaves erratically out of vocabulary: zero-adaptation accuracy across four cross-dataset directions fluctuates between 7.35% and 96.40%, and under out-of-vocabulary weights its confidence fails inversely on open-set evaluation, assigning higher confidence to unseen than to seen classes (AUROC 0.38), a failure we term silent hallucination. We retain the slot representation and replace the readout with RSAJEV, a training-free abstaining prototype memory that performs retrieval-based prototype matching: K reference images per class are cached as prototypes via their slot representations under the same prompt, and classification becomes nearest-neighbor retrieval over the prototypes, refined by a shrinkage-Mahalanobis…

    PDF
  • arXiv cs.AIOct 11

    Visible but Not Avoided: State-Dependent Re-Proposal of Rejected Candidates in Catalog-Based LLM Optimization Loops

    Jianlin Chen, Wenhui Chen, Chi Man Vong

    LLM optimization loops often show the proposer the candidates it has already tried and rejected, but leave it to the proposer to avoid them. We ask whether avoidance measured in a static probe predicts avoidance in the states an optimizer's own trajectory produces. In catalog-based loops where three open LLMs improve a GUI agent by naming one of 40 edits, a re-proposal is identified exactly by the edit's name. Clean base-state probes underpredict re-proposal in the proposers' own loops (macro-averaged loop-minus-probe rate 0.32 [0.27, 0.37]), and proposals of already evaluated harness-edit pairs consume 28-63% of valid proposer iterations. The two settings reject different edits: a probe's rejected edits are mostly ones the proposer did not choose, whereas a loop's are its own earlier choices, which remain preferred when the history is removed. Ordinary history lowers re-proposal for two proposers but raises it for the third, and the effects of failure evidence are model-specific. Stating the rejection status in one sentence lowers re-proposal in every model (-0.16 [-0.22, -0.09] relative to a neutral list of the same IDs). For catalog-based loops with exact edit identities,…

    PDF
  • arXiv cs.CVOct 11

    ScenePlacer: Reformulating Scene-Level Object Pose Estimation as Visibility Field Learning

    Ken Deng, Jiarui Liu, Guibiao Liao, Kunming Luo, Mochu Xiang +5

    We present ScenePlacer, a framework for scene-level pose estimation of objects generated by state-of-the-art native shape generation models, placing generated 3D objects into scenes given a partial geometric observation without modifying the underlying generative pipeline. Existing scene generation approaches either generate objects directly in scene coordinates, compromising their canonical-space priors, or regress poses from complete generated shapes in the shape generation process - the latter being a poorly conditioned problem that is sensitive to scale ambiguity and partial observations. ScenePlacer decouples object generation from pose estimation by reformulating alignment as conditional visibility-field learning. Given an observed partial point cloud from reconstruction and a generated mesh, it predicts a visibility field defined in object space, then estimates the rigid transformation by aligning the partial observation to visible parts on the mesh. We propose a pretrained 3D shape VAE as our backbone, preserving its mesh prior while conditioning it on a separately encoded observation. We further introduce visibility-aware, radius-constrained supervision and synthesize…

    PDF
  • arXiv cs.CLOct 11

    NepPlural: A Human-Anchored Benchmark for Persona Annotation in Nepali Migration Discourse

    Jinu Nyachhyon, Anish Khatiwada, Bimal Timilsina

    Pluralistic alignment asks that language models faithfully represent a society's diverse, competing values rather than collapse them into one dominant viewpoint, yet flagship pluralistic-alignment datasets remain heavily skewed toward Western respondents. We introduce NepPlural, a human-anchored benchmark for modelling competing socio-cultural personas in Nepali public discourse on youth migration and brain drain. We curate 2,148 code-mixed (Devanagari, Romanized Nepali, English) YouTube comments, annotate them along a four-axis persona taxonomy (Intent, Primary Driver, Value Orientation, Affect) plus an explicit No-Persona class, and build a 300-comment gold set labelled independently by three native Nepali speakers (mean Fleiss' $κ\approx 0.55$). Using this gold set to audit automatic annotation, we find that LLM annotation reliability is driven by protocol. Labels collected in batches through chat interfaces reach only near-chance inter-model agreement ($κ\approx 0.17$). Annotating one comment at a time raises agreement above the native-speaker level ($κ\approx 0.72$), but consistency is not validity: under a forced-choice schema, models agree on personas for the 39% of…

    PDF
  • arXiv cs.LGOct 11

    A Dependency-Aware Multi-GPU Training Framework for Temporal Graph Neural Networks

    Yang Zhao, Yue Dai, Liang Qiao, Buxin Tu, Bo Liu +6

    Temporal graph neural networks (TGNNs) achieve superior performance in many dynamic graph learning tasks. These models use node state vectors (i.e., memories) to capture and record graph changes (i.e., events), and then exploit these memories to assist prediction. However, this memory-centric design makes multi-GPU training challenging because predictions for later events often depend on memory updates from earlier events. To maintain model quality, existing multi-GPU TGNN training schemes strictly preserve all event dependencies, which limits parallelism and causes expensive cross-GPU memory synchronization. In this work, we observe that many of these dependencies can be handled more flexibly without harming training quality. Specifically, (a) dependencies in TGNN training have high locality in both time and graph topology, (b) the short-range dependencies missed within a large batch can often be recovered cheaply, and (c) many memory updates are not immediately needed. Based on these observations, we propose DepBloom to accelerate multi-GPU TGNN training with three techniques: (1) temporally bounded spatial partitioning to reduce unnecessary cross-GPU synchronization and…

    PDF
  • arXiv cs.CVOct 11

    Visually Legible but Adversarial: Watermark-Constrained Attacks on End-to-End VLMs for Document Understanding

    Qingzhe Han, Fan Dang, Shengqi Chen, Xinyi Sun, Haoyu Xiong +2

    Visible watermarks convey document ownership, confidentiality, and status, but can also carry adversarial signals. We introduce Auto-Regional Adversarial Watermarking (AR-AWM), a model-aware white-box attack on end-to-end visual language models (VLMs) for document understanding. AR-AWM constructs a small, spatially distributed watermark-text support and optimizes only its pixels to disrupt visual representations. The attack is untargeted and question-agnostic: generation uses no questions, answers, parsing prompts, language-model gradients, or language-model outputs. Each attacked page is reused across downstream queries. On Qwen2.5-VL-7B, a watermark support covering 7.98% of page pixels under the reporting threshold reduces DocVQA exact match from 0.899 to 0.150 and average normalized Levenshtein similarity (ANLS) from 0.939 to 0.212. Selected qualitative cases retain visually legible source fields. The same support without optimization and with matched random perturbations retains ANLS of 0.932 and 0.933, respectively, identifying adversarial optimization as the primary cause of degradation. Layout-matched optimized rectangles yield ANLS of 0.276, showing an incremental…

    PDF
Load more
PDF
PDF