The paper investigates how large audio‑language models (LALMs) hallucinate objects by affirming sounds that are not present. Using a black‑box causal test, the authors isolate evidence regions via CLAP similarity, remove them with spectral gating, and compare against random windows across a 2×2×2 grid on three LALMs. The study finds that evidence localization improves detection over random removal, that raw detection scores are model‑dependent, and that a contrastive decoding strategy combining blank and language‑prior cues boosts Qwen2‑Audio accuracy from 57.1 % to 71.9 %, outperforming Audio‑Aware Decoding alone.
KV-Repurpose is an open‑source, vLLM‑based system that enables segment‑wise KV‑cache reuse beyond exact prompt prefixes in large language model serving. It integrates cache‑aware segment scheduling, prompt construction, capacity management, workload generation, and runtime monitoring. Experiments show that while naive segment reuse maintains output quality for single‑document tasks, it reveals limitations in cross‑document reasoning, and under a Zipfian RAG‑style workload it achieves a 63.9% token‑level cache hit rate versus 3.3% for prefix caching at 30 GB, dramatically lowering time‑to‑first‑token.
CliQ‑RRG is a two‑stage framework for generating radiology reports that aligns visual and textual data using disease‑aware embeddings and a Prior‑Guided Attention Module to capture multi‑view dependencies. In the second stage, GraphRAG‑guided retrieval adds graph‑grounded clinical evidence, and a large language model restructures the content into concise question‑answer pairs with diagnostic summaries. Experiments on two chest X‑ray benchmarks show that CliQ‑RRG outperforms previous methods in both lexical and clinical metrics, producing accurate and clinically coherent QA‑style reports.
The paper investigates how different visual representations and integration mechanisms affect the performance of World Action Models (WAMs). It finds that strong general‑purpose visual representations yield the largest robustness gains, that channel‑wise joint prediction offers the best success rate, and that additional sources should only be added if they contribute task‑relevant information beyond existing representations. The authors propose a Representation Recipe (R$^2$‑WAM) that achieves state‑of‑the‑art success rates on RoboCasa GR1 Tabletop and RoboTwin 2.0.
The paper introduces Finsler Flow Matching (FFM), a method for learning continuous stochastic dynamics from discrete Markov transition graphs in single‑cell snapshot data. FFM constructs a Finsler geometry using first and second local moments, learns directed geodesics via neural networks, and employs these to build geometry‑aware stochastic paths that can be simulated without explicit score or flow matching. Benchmark tests show that FFM better recovers intermediate cellular states, especially when dynamics are directional or anisotropic.
By Niklas Canova, Jonas Simon Fleck
The paper investigates the relationship between large language models (LLMs) and compression, showing that LLMs, modeled as next‑token predictors, are equivalent to monotone (order‑preserving) compression algorithms with an error gap of at most 2. It further proves that this monotonicity is necessary for the equivalence only if cryptographic one‑way functions exist, and derives a cryptographic corollary linking next‑bit pseudoentropy to monotone incompressibility.
By Noam Mazor, Andrew Morgan, Rafael Pass
The paper introduces Policy Alignment Membership Auditing (PAMA), a framework for detecting whether a prompt was used during on‑policy distillation (OPD). PAMA leverages the direction of teacher‑guided policy updates, quantified by Teacher Alignment Gain (TAG), alongside student drift and uncertainty signals to assess membership. Experiments on six datasets, including the MATH benchmark, show that PAMA improves AUC by 14.6–20.6% over existing methods.
By Yilong Yang, Wenzhuo Shang, Yule Liu, Jiale Teng, Zhuo Ma
WOVEN is a new training source and benchmark for visual transition reasoning, comprising 36,076 examples across 20 scenes, 5 actions, and 8 reasoning types. The study shows that current multimodal large language models (MLLMs) have a systematic deficit in spatial, embodied, physical, and temporal reasoning, and that training on WOVEN improves performance on 22 of 26 external benchmarks by up to 27.3 percentage points. The authors also provide a training recipe that prioritizes reasoning operations and larger visual changes for robust visual world modeling.
By Zheyu Fan, Yue Zhang, Mingkai Deng, Kangrui Wang, Qineng Wang, Canyu Chen, Jie Hao, Xing Fan, Chenlei Guo, Eric P. Xing, Mohit Bansal, Manling Li
The paper argues that sparse attention should be viewed as a matrix approximation problem rather than merely selecting large scalar entries from the attention matrix. It introduces Matrix Approximation Sparse Attention (MASA), a method that replaces raw attention-mass ranking with a closed‑form score reflecting the reduction in matrix‑product approximation error. MASA can be integrated into existing sparse attention frameworks and, according to extensive experiments, consistently improves accuracy across various methods, benchmarks, and LLM backbones.
By Fang Wan, Xufeng Liu, Fan Li, Yi Liu
The paper introduces Uncertainty-Aware Budget Allocation (UAB), a concave integer optimization framework that reallocates a fixed sampling budget for language model reasoning based on uncertainty estimated from initial samples. In Phase‑1, each question receives a small number of generations, and the disagreement among these answers (vote entropy) signals difficulty. Phase‑2 uses a marginal‑greedy algorithm to allocate the remaining budget, focusing on questions with high initial disagreement, leading to a 2.3% average accuracy improvement over uniform allocation across five open‑weight models and five reasoning benchmarks, with up to 5.5% gains on individual benchmarks, especially in low‑resource settings.
By Manh Nguyen, Sunil Gupta, Hung Le
The paper introduces Uni-DSM, a unified score-driven framework that uses denoising score matching to model video anomaly patterns for both detection and anticipation. It employs a shared noise-conditioned score transformer backbone with scene-dependent embeddings and motion-aware weighting, and differentiates between VAD and VAA through autoregressive denoising and a self-distilled denoising mechanism. Experiments on multiple benchmarks show state‑of‑the‑art performance for both tasks while maintaining high efficiency.
By Congqi Cao, Zhenhe Liang, Hanwen Zhang, Yifan Zhao, Qinyi Lv, Lingtong Min, Yanning Zhang
EvoRubric introduces a co‑evolutionary reinforcement learning framework that learns rubric‑driven evaluation criteria for open‑ended text generation. The system uses a shared policy that simultaneously generates responses and rubrics, while a frozen meta‑verifier and grader provide validity checks and scoring. Experiments on medical, writing, and science benchmarks show that EvoRubric outperforms static and other evolving‑rubric baselines, achieving higher average scores on five tasks.
By Xin Guan, Xiaomeng Hu, Shen Huang, Zhenyi Wang, Bo Zhang, Zijian Li, Pengjun Xie, Bo Liu, Jiuxin Cao
3DTexMOR introduces a method for removing multiple 3D objects from reconstructed scenes by performing texture-space inpainting rather than 2D image inpainting. The approach aggregates observations into shared texture maps, inpaints missing regions, and reprojects the completed maps to supervise Gaussian scene completion, while decomposing appearance to avoid view-dependent artifacts. Experiments show that this technique yields visually plausible results and outperforms existing methods, improving PSNR by 5.8 dB and reducing LPIPS by at least 22%.
By Kunxin Guang, Yonghao Zhao, Jian Yang, Beibei Wang
The paper presents TopoGPU, a GPU streaming pipeline that computes persistence diagrams of cubical complexes from 2D and 3D images. It processes image chunks with massively parallel GPU kernels, accumulates boundary relations in host memory, and performs boundary matrix reduction on the CPU. TopoGPU introduces a stratification‑aware discrete Morse matching, parallel topological sorting, and a parallel V‑path parity algorithm, achieving up to 198× speedup over Cubical Ripser and reducing persistent homology computation costs in a topology‑preserving deep network.
By Fan Wang, Hubert Wagner, Rezaul Chowdhury, Chao Chen
The paper introduces ResOT, a training‑free method that mitigates object hallucination in large vision‑language models by repairing hidden representations at inference time. ResOT aligns hallucinated directions with faithful ones through localized distribution alignment and Gaussian optimal transport, creating a low‑dimensional residual subspace for minimal intervention. Experiments on three LVLMs demonstrate significant reductions in hallucination and improvements in image caption quality and multimodal performance.
By Chen Zhao, Xingping Dong, Jiachun Shi, Liang Peng, Chong Wang, Zhen Lei, Ran He, Bo Du
GLM-RAG introduces a graph language model (GLM) based retriever for retrieval‑augmented generation over knowledge graphs. The study compares GLM‑based, graph neural network (GNN)‑based, and vector‑search retrievers across single‑ and multi‑hop settings, emphasizing transferability to unseen domains. Results show that finetuned GLM retrievers achieve state‑of‑the‑art performance on two multi‑hop benchmarks and generalize better out of domain, while GNN retrievers excel at graph coverage and vector‑search performs best on single‑hop tasks.
By Maya Arseven, Anette Frank, Beni Egressy, Johann Higl, Moritz Plenz
This survey reviews industrial anomaly synthesis (IAS), a field that has been underrepresented in prior studies focused on anomaly detection. It categorizes IAS methods into four paradigms—hand‑crafted, distribution‑hypothesis, generative model, and vision‑language model—while presenting a taxonomy that facilitates systematic comparison. The paper also discusses datasets, benchmarks, evaluation metrics, multimodal advances, and practical deployment considerations, offering guidance for future research.
By Yanshu Wang, Xichen Xu, Jinbao Wang, Chengbin Ma, Xiaoning Lei, Guoyang Xie, Guannan Jiang, Zhichao Lu
The paper introduces Interventional Transfer (IT), a method for evaluating rubric generation by assessing how two rubrics respond to perturbations that cause a pass/fail shift. IT reveals asymmetries in rubrics generated by different LLMs when applied to HealthBench, showing that improvements on one rubric do not reliably transfer to the other. The study contrasts IT with existing evaluation approaches, highlighting its ability to uncover these asymmetries.
By Erik Skalnes, Layne C. Price, Raviteja Anantha, Michael Oberst
FeatCal is a post‑merging calibration method that reduces feature drift between a merged model and its task experts by adjusting weights layer by layer in forward order using a small calibration set. It employs a closed‑form update that avoids gradient descent or iterative optimization. On CLIP and GLUE benchmarks, FeatCal outperforms existing calibration baselines, achieving higher accuracy with fewer calibration examples and lower computational cost.
By Yanggan Gu, Shuo Cai, Zihao Wang, Wenjun Wang, Yuanyi Wang, Pengkai Wang, Sirui Huang, Su Lu, Jianmin Wu, Hongxia Yang
The paper introduces the concept of a prior barrier to measure how well pretrained large language models support different concepts during supervised fine‑tuning (SFT). It shows that these barriers follow a long‑tail distribution, causing head concepts to be easier to learn than tail concepts. To address this, the authors propose PASS, an adaptive instruction‑selection method that allocates limited supervision budget toward under‑covered concepts, outperforming seven state‑of‑the‑art methods across multiple settings.
By Haohui Wang, Jiahao Xu, Wangzhi Zhan, Tong Zeng, Dongqi Fu, Hong Li, Swastik Roy, Naren Ramakrishnan, Chris North, Jian Kang, Yujun Yan, Dawei Zhou