ReactHuman is a physics‑grounded benchmark that tests whether multimodal large language models (MLLMs) can make immediate, safety‑critical decisions in simulated humanoid scenarios involving sudden household hazards. The benchmark includes 17 event families, over 1,000 reproducible scenes generated from 240 Hz rigid‑body simulation, and a five‑metric suite evaluating reactions on reasonableness, safety, and physical grounding. Evaluation of seven MLLMs reveals that reactive safety remains unsolved, with models frequently mishandling hazards, relying on appearance over motion, and missing key interception points.
By Yizhan Li, Jianxin You, Mengyang Xiong, Yinhuan Chen, Zicheng Zhao, Dekun Wu, Dongqing Zhang, Bang Liu
X-AuT is a progressive framework for compressing the audio encoder of speech large language models. It selects layer combinations via short behavioral probes and restores performance through representation alignment, cross‑scale distillation, scheduled student‑policy supervision, and LoRA finetuning, while keeping the language‑model backbone frozen. On ten Chinese–English benchmarks, reducing Qwen3‑ASR‑0.6B’s encoder from 18 to 16 layers lowers macro‑average error from 5.61% to 5.27%, and a 14‑layer model achieves 5.75% error with 20.7% fewer parameters.
By Haojun Zhang, Yi Zou, Min Chen, Qize Yu, Lianrui Fan, Xini Ding, Hao Li, Shuchang Zhou, Xianming Liu, Shiyu Huang
RAMamba-Net is a new multimodal fusion network designed for auditory attention decoding (AAD) that combines EEG and electrooculography (EOG) signals. It uses a Mamba-enhanced band-aware convolutional Transformer to capture EEG band-specific patterns and long-range temporal dynamics, while a dual-branch encoder models EOG temporal and inter-channel dependencies. Cross‑modal attention and a reliability‑aware module estimate sample‑wise modality weights, improving fusion robustness and achieving a 5.76% accuracy gain over unimodal baselines on two AAD benchmarks.
By Xingyi He, Ziwei Wang, Dongrui Wu
Mr.LHDR is a new benchmark designed to evaluate deep research agents on long‑horizon, multimodal tasks. It presents questions built from hidden Node‑Relation graphs that require an average of 12.1 intermediate conclusions and a mean dependency depth of 10.4 before arriving at a single verifiable answer. The benchmark tests both final answers and the correctness of intermediate conclusions, using metrics such as Overall Accuracy, Strict Accuracy, Checklist Score, and Dependency‑Aware Checklist Score.
By Minghao Guo, Meng Cao, Sui Zhao, Siyu Ning, Xin Wang, Haoze Zhao, Jiaxuan Yang, Haihong Hao, Mingfei Han, Shunlin Rong, Haijun Wu, Xiaodan Liang, Xiaojun Chang
The paper introduces a routing-by-reasoning-need controller for diffusion vision‑language models that dynamically selects early commitment, baseline preservation, or reasoning‑supportive decoding based on trajectory signals such as answer closure, commitment evidence, and representation revision pressure. This training‑free approach avoids a single universal generation length, instead tailoring inference‑time control to the reasoning demands of each question. Experiments on answer‑focused, mixed‑reasoning, and chain‑of‑thought benchmarks show that routed control improves robustness compared to fixed long or short decoding and single‑rule interventions, with benefits not solely due to shorter outputs.
By Yixiang Liu, Zhongxing Xu, Zhonghua Wang, Xiaoying Tang
arXiv:2609. 11697v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) and World-Action Models (WAMs) have demonstrated strong capabilities in general-purpose robotic manipulation, yet their generated actions may violate hard physical constraints and therefore be unsafe or infeasible for deployment.
By Jianming Ma, Rongjun Jin, Xiaxi Si, Yang Zhang, Yiheng Li, Yue Gao
arXiv:2609. 11916v1 Announce Type: new Abstract: Camera traps often run in the field on edge hardware with limited or no connectivity, making small, locally-deployable vision-language models (VLMs) -- not frontier-scale ones -- the practically relevant class to evaluate for species identification.
By William Zhou, Mayukha Siripuram, Xiao Yan, Ziqi Liu, Yi Ding
Sci‑MMR is a new benchmark for multi‑step evidence‑grounded scientific reasoning in multimodal agents, featuring 235 multi‑hop tasks across four disciplines and an average of nine figure panels per task. It evaluates not just final answer accuracy but also the recovery of structured evidence from scientific claims, citations, visual data, and supporting regions. Experiments on eight state‑of‑the‑art models show a gap of over 20 points between answer accuracy and complete evidence recovery, highlighting significant challenges in evidence acquisition and integration.
By Jiaqiang Li, Yajie Yang, Zhiheng Xi, Jiadong Chen, Enyu Zhou, Senjie Jin, Yang Nan, Jiazheng Zhang, Han Wang, Yanxin Li, Dingwei Zhu, Bicheng Deng, Yuhui Wang, Xiang Zheng, Qi Zhang, Lei Bai, Xingjun Ma, Tao Gui
The paper introduces a continuous‑time approach to acoustic modeling in text‑to‑speech (TTS) using neural controlled differential equations (CDEs). It treats phone representations as temporally parameterized control paths and employs a neural acoustic vector field to generate hidden states that evolve continuously with phonetic content and duration. Experiments show that CDE‑based models can improve emotion‑intensity ranking while preserving expression quality, and that adjusting temporal resolution affects style tracking versus calibration.
By Mattias Cross, Minghui Zhao, Anton Ragni
The paper introduces a method for measuring accent differences that balances interpretability and practicality. It proposes using articulatory representations obtained via articulatory inversion as an interpretable basis for accent comparison, while employing optimal transport to compare accents across any type of recording. This approach aims to overcome the limitations of traditional phonetic analyses and embedding‑based methods, which are either time‑consuming or non‑interpretable.
By Charles McGhee, Mark J. F. Gales, Kate M. Knill
The paper introduces a synthetic benchmark for multimodal time‑series forecasting that evaluates how well text annotations contribute to predictions. By generating controlled signals with semantically correct, incorrect, and irrelevant annotations, the authors can precisely measure the true information content. Six mutual‑information estimators (KSG, MINE, InfoNCE, CCA, PID, and V‑information) are tested, all correctly ranking useful annotations and enabling annotation auditing without model training. The benchmark also highlights each estimator’s limitations and validates findings on seven real datasets, providing practical guidelines for metric implementation.
By Emma Andrews, Gianmarco Mengaldo
The paper introduces 2AM, a system that separates memory and action execution in long‑horizon robot manipulation. 2AM stores task memory exclusively in a multimodal Agent, while a single RGB‑based, stateless Action Model performs motion based on language and optional 2D hints. On the LIBERO‑Mem benchmark, 2AM achieves 76.3% average completion—over 61 points higher than the best baseline—demonstrating that agent‑side memory and precise steering of the Action Model can substantially improve performance.
By Yutong Hu, Fengjiao Chen, Xuezhi Cao, Renaud Detry
The paper introduces CoMA-DiT, a bidirectional cross‑modal Diffusion Transformer that uses paired modalities as mutual generative supervision for latent augmentation rather than just inputs for fusion. By conditioning velocity prediction on the paired modality through cross‑modal attention and injecting variation via a reliability‑gated residual mechanism, CoMA‑DiT improves multimodal brain state decoding. Experiments on auditory attention decoding and emotion recognition show consistent gains over 20 baselines, with absolute accuracy and macro‑F1 improvements of 4.28% and 6.70% respectively, and extensive analyses confirm its robustness and interpretability.
By Ziwei Wang, Xingyi He, Hongbin Wang, Tianwang Jia, Bohan Fang, Dongrui Wu
The paper investigates how large language models (LLMs) interpret ambiguous or incomplete text prompts for visualization authoring and introduces visual prompts as a complementary modality to improve precision. An empirical study informs the design of VisPilot, a system that allows users to create visualizations using text, sketches, and direct manipulation. A controlled user study and expert evaluation show that multimodal prompts help users convey spatial constraints, local references, and design preferences while maintaining task efficiency comparable to text-only prompting.
By Zhen Wen, Luoxuan Weng, Yinghao Tang, Runjin Zhang, Yuxin Liu, Bo Pan, Minfeng Zhu, Wei Chen
The paper introduces the Spectral Masking and Interpolation Attack (SMIA), a black‑box adversarial technique that subtly alters inaudible frequency regions of AI‑generated audio to fool voice authentication systems and their countermeasures. Experiments show SMIA achieves at least 82% success against combined verification and countermeasure systems, 97.5% against standalone speaker verification, and 100% against countermeasures, revealing a critical security gap. The authors argue that current static defenses are inadequate and call for dynamic, context‑aware defenses that can adapt to evolving threats.
By Kamel Kamel, Hridoy Sankar Dutta, Keshav Sood, Sunil Aryal
The paper presents a deep‑learning approach for detecting cyberattacks in Low‑Earth Orbit satellite systems, leveraging the UNSW‑IoTSAT dataset. It explores structured architectures that preserve hardware, orbital, and radio‑frequency data, including a Subsystem‑Fusion MLP and a hierarchical multimodal Transformer that captures cross‑subsystem interactions and temporal dynamics. Experiments show that the hierarchical Transformer achieves up to 91.66% accuracy and 85.63% macro F1 under a leakage‑resistant evaluation protocol, highlighting the importance of multimodal modeling and rigorous testing.
By Kyle Stein, Guillermo Francia III, Eman El-Sheikh, Hossain Shahriar
The paper introduces Time‑Frequency Geometric Cross‑Attention (TFGCA), a module that enhances vision‑language‑action models by decomposing action chunks into time‑frequency tokens using a learnable wavelet transform. TFGCA fuses dot‑product similarity with wedge‑product magnitude to better capture both frequency‑based smooth trends and cross‑phase orthogonal motion structures. When added to a pretrained VLA model, it yields significant performance gains across in‑distribution and out‑of‑distribution benchmarks, including a 28.5‑point improvement under RoboTwin domain randomization and an 11.67‑point increase on real‑robot AgiBot A2 tasks.
By Shengye Dong, Haochen Niu, Hao Liu, Peiwen Lin, Chuang Wang, Shanmin Pang
The paper introduces a framework to evaluate and diagnose the robustness of low‑resource multilingual text‑to‑speech systems when faced with complex text inputs such as numbers, dates, named entities, long sentences, code‑switched expressions, and punctuation structures. It assesses robustness across content consistency, language consistency, and generation stability, and proposes automatic metrics (character error rate, language ID accuracy, duration abnormal rate) along with a lightweight Text Risk Score (TRS) that predicts synthesis risk from interpretable text features. Experiments on Thai, Vietnamese, Swahili, and Indonesian TTS systems reveal distinct failure patterns and show that TRS correlates positively with content and duration errors, offering a low‑cost pre‑synthesis risk indicator.
By Tianlun Zuo, Ziyu Zhang, Tingzhi Mao, Zhonghua Fu, Lei Xie
The Eloquence team presents three methods for the Interspeech 2026 MLC‑SLM Task 2, a multilingual MCQA challenge covering 21 languages. They fine‑tune Voxtral‑Mini‑3B with LoRA and data augmentation, achieving 0.72 macro‑accuracy; they use multimodal in‑context learning on Voxtral‑24B to correct label bias, reaching 0.81; and they deploy a training‑free retrieval system with a voice‑anchored memory, scoring 0.68. All approaches surpass the official baseline.
By Jordi Luque, Lorenzo Concina, Marco Matassoni, Alessio Brutti, Filippo Vella
OmniKVQuant introduces a training‑free framework for quantizing the key‑value (KV) cache of omni‑modal large language models (Omni‑LLMs) that process audio, video, and text simultaneously. The method addresses two identified problems—temporal key drift and heterogeneous value geometry—by setting key quantization ranges over short input windows and rotating values separately for each modality. Applied to Qwen2.5‑Omni and Qwen3‑Omni, OmniKVQuant achieves 2‑bit KV caches while largely preserving performance across seven audio‑visual benchmarks, and includes a fused Triton decode kernel that eliminates the need for a dense FP16 cache.
By Suho Yoo, Hyunjong Ok, Jongmin Choi, Jihoo Jung, Joon Son Chung