The paper introduces SMILE, a self‑explainable multimodal information bottleneck framework for medical diagnosis. It jointly optimizes predictive accuracy and modality‑specific explainability by selecting the most informative elements within each data modality. Experiments on diverse medical datasets show strong diagnostic performance, including a 9.1‑percentage‑point accuracy gain on the iCTCF dataset, and provide transparent, modality‑aware explanations that enhance both explainability and generalization.
By Yuqing Yang, Alexander Schmatz, Zhaozhao Ma, Changkyu Choi, Robert Jenssen, Shujian Yu
MedProb is a lightweight probing framework that predicts multiple-choice medical visual question answering (Med‑VQA) answers directly from frozen vision‑language model (VLM) representations, avoiding free‑text generation. On datasets such as PATH‑VQA, SLAKE, and VQA‑RAD, MedProb extracts more answer‑relevant signal than prompting and outperforms both medical VLMs and agentic systems. The approach also narrows the performance gap between small and large models, shows that medical adaptation does not consistently improve linear decodability, and reveals positional biases in both prompting and generation.
By Erfan Nourbakhsh, Ke Yang, Anthony Rios
TRILOGUE is a new trilingual benchmark for spoken dialogue fact‑checking, covering English, Russian, and Kazakh. It includes almost 12,000 dialogues, 187,000 turns, and 390 hours of paired audio with ASR transcripts and word‑level timestamps, as well as nearly 5,000 human‑recorded Russian and Kazakh files. The dataset supports tasks such as claim check‑worthiness detection, evidence retrieval, and claim verification under various input conditions, and baseline experiments reveal challenges with ASR errors and cross‑lingual transfer, especially for Kazakh.
By Chaewan Chun, Meruyert Aristombayeva, Jiyoung Choi, Mahjabin Nahar, Delvin Ce Zhang, Dongwon Lee
BioSync is a transformer-based model that fuses cardiac, neural, behavioral, and speech data from wearables and mobile devices into a continuous composite digital biomarker called the BioSync Index (BSI). The architecture uses multi-head self-attention on modality tokens and a linear branch for feature concatenation, inspired by latent-variable measurement theory. Evaluations on synthetic cohorts for cognitive decline and metabolic-autonomic conditions show BioSync achieving AUCs of 0.928 and 0.764 accuracy/F1 of 0.766, outperforming simple concatenation and other fusion strategies in most corruption scenarios.
By Seyed Mahmoud Sajjadi Mohammadabadi
The paper introduces a training‑free, inference‑time technique to curb hallucinated transcripts in Whisper by projecting decoder activations away from a low‑rank hallucination subspace derived from non‑speech data. Two variants are tested: an always‑on projection that dramatically lowers hallucination rates on non‑speech benchmarks, and a gated version that applies the projection only when non‑speech is predicted, achieving a smaller but still significant reduction. On LibriSpeech, the gated method slightly increases word error rate but keeps false‑rejection rates low, demonstrating a controllable trade‑off between hallucination suppression and recognition accuracy.
By Maryam Abbasihafshejani, Murtuza Jadliwala
ElderBench is a new benchmark designed to evaluate autonomous mobile agents in scenarios that reflect the natural language patterns of older adults. It comprises 249 smartphone tasks collected from older users across 20 applications, highlighting linguistic differences such as indirect speech, referential ambiguity, and under-specified requests compared to existing GUI benchmarks. The study shows that mainstream GUI agents and Vision‑Language Models perform significantly worse on these elderly‑oriented instructions, and it offers insights into how specific language patterns lead to failures.
By Weide Zhan, Qumu Shaqu, Yuanqing Liu, Peng Zhang, Jiahao Liu, Kam Him Lam, Ning Gu, Zhan Hu, Tun Lu
MM-IFEval-Pro is a new multilingual benchmark for evaluating instruction-following in vision-language models, covering both Chinese and English tasks. It includes 4 major task categories, 24 subcategories, and 8 instruction categories with 52 subcategories, each sample featuring an average of 3.0 constraints to mimic complex instruction scenarios. A reinforcement-learning training set with Chinese and adversarial instructions improves model performance on MM-IFEval-Pro and transfers well to other multimodal benchmarks, showing strong cross-task and cross-language generalization.
By Changming Xiao, Zhenliang Ni, Jinhui He, Han Shu, Jie Hu
ProCA: Progressive Contrastive Alignment for Robust EEG Visual Decoding introduces a model‑agnostic framework that adaptively aligns EEG signals with visual semantics. It replaces fixed visual or textual anchors with EEG‑aware class‑level contrastive supervision and employs structure‑consistent interpolation to preserve channel‑wise and temporal importance. Across multiple evaluation settings—including subject‑dependent, subject‑independent, strict cross‑subject transfer, and continual adaptation—ProCA delivers significant performance gains, achieving relative Top‑1 improvements ranging from 7.4% to 28.1%.
By Kanglei Zhou, Chunyan Lan, Dongyang Li, Jun Zhu, Liyuan Wang
The paper surveys how commonsense reasoning is being integrated into computer vision, moving beyond traditional CNNs that only detect objects. It reviews methods that use knowledge graphs, scene graphs, neuro-symbolic models, and transformers to add contextual understanding, thereby improving object recognition and spatial reasoning. The authors also discuss current limitations such as dataset bias and knowledge gaps, and propose future research directions in cross‑modal reasoning, scalable knowledge injection, and hybrid architectures.
By Bahar Uddin Mahmud, Sumit Barua, Guan Yue Hong, Ajay Gupta, Hexu Liu
The paper introduces the Cross‑Modal Triage Network (CMTN), a multimodal deep‑learning model that fuses a Swin Transformer V2 visual encoder with a PubMedBERT text encoder to perform severity‑based triage, pathology detection, and generate visual explanations for chest radiographs. Trained on 34,639 image‑text pairs from MIMIC‑CXR‑JPG, the CMTN achieves high ordinal agreement with reference labels (QWK = 0.9341) and excellent pathology detection (macro‑AUROC = 0.9970) while operating with 34 ms latency. However, a blinded clinical audit revealed low agreement with expert radiologists (QWK = 0.1399) and only modest spatial‑semantic concordance in heatmaps, underscoring the gap between algorithmic performance and clinical judgment.
By Zinah Ghulam, Richa Mittal, Eranga Ukwatta
arXiv:2609. 04552v1 Announce Type: cross Abstract: Unattended interactive autonomy - machines that step into danger in place of humans and complete tasks with human tools - remains a missing capability in mission-critical operations.
By Amarjot Singh, Tanmay R. Pancholi, Jainam Kothari, Shrirang Mahajan, Ketan Bansal, Zackory Erickson, Giuseppe Loianno, Alexandre M. Bayen, Jeff Schneider, Vince Nakayama
The paper introduces a multimodal emotion recognition framework that combines audio and visual feature extraction with an attention-based fusion strategy. Audio features include Wav2Vec2 embeddings, MFCCs, and statistical acoustic descriptors, fused via a BiLSTM, while video features are extracted using a ResNet50-BiLSTM architecture. A multi-head attention mechanism fuses these modalities, and experiments on MELD and IEMOCAP show significant accuracy and robustness gains, especially in unbalanced data settings.
By Xu Lin, Ke Wang, Hui Kang, Xinying Wang
PRISM‑Bench is an audio‑centric diagnostic benchmark for text‑to‑audio‑video generation, built from 900 human‑verified samples. It evaluates audio along two axes—audio type (speech, music, sound) and sound‑source visibility (on‑screen vs. off‑screen)—across four perceptual dimensions (audio‑visual coherence, audio quality, audio expressiveness, and prompt following) using 35 fine‑grained criteria. The benchmark employs an enhanced MLLM‑as‑a‑Judge protocol that aligns strongly with human raters, revealing a performance gap between frontier and open‑source T2AV models and highlighting overfitting to perceptual fidelity while struggling with complex grounding and control tasks, especially for music and synchronized on‑screen audio.
By Yuchen Sun, Qian Yang, Jun Wang, Detai Xin, Guoqiao Yu, Guanglu Wan, Qi Jia
MCPO introduces a two-stage method to compress multimodal chain-of-thought reasoning. The first stage uses a step-level Normalized Cross-Modal Mutual Information pruning algorithm to remove visual-independent steps, reducing redundancy and hallucinations. The second stage fine-tunes the model with an asymmetric multimodal length-controlled preference loss, achieving up to 69.5% shorter reasoning chains and a 3.34× inference speedup while keeping accuracy.
By Guangheng Yang, Zhenliang Ni, Zhenkai Wu, Han Shu, Juan Feng, Wenming Yang, Jie Hu
RL-VLA$^3$ is a fully asynchronous distributed reinforcement learning framework designed for Vision‑Language‑Action (VLA) model training. It allows fine‑grained asynchronous interaction between simulation, inference, and training via dynamic batching schedulers and flexible environment sharding, addressing the variable, resource‑intensive latencies of physical simulators. Experiments across multiple simulation backends, VLA architectures, and RL algorithms show throughput gains of up to 85.2% over synchronous baselines while preserving sample efficiency, and the system scales from 8 to 256 GPUs.
By Haoran Sun, Yongjian Guo, Zhong Guan, Shuai Di, Xiaodong Bai, Jing Long, Tianyun Zhao, Mingxi Luo, Hongke Zhao, Likang Wu, Xiaotie Deng, Xu Chu, Xi Xiao, Sheng Wen, Yicheng Gong, Junwu Xiong
The paper introduces Expert Modulation, a novel approach for multi‑modal time series prediction that conditions both expert routing and computation on textual signals, thereby providing direct cross‑modal control over expert behavior. Unlike previous methods that rely on token‑level fusion, this mechanism avoids mixing temporal patches with language tokens in a shared embedding space, which can be problematic when high‑quality time‑text pairs are scarce or when time series characteristics vary widely. Experiments and theoretical analysis demonstrate that Expert Modulation yields strong improvements over existing multi‑modal forecasting techniques.
By Lige Zhang, Ali Maatouk, Jialin Chen, Karthik Charan Konduri, Leandros Tassiulas, Rex Ying
The paper introduces Post Fusion Stabilizer (PFS), a lightweight module that refines intermediate bird’s‑eye view (BEV) feature maps in existing camera‑LiDAR fusion detectors. PFS stabilizes feature statistics under domain shift, suppresses regions affected by sensor degradation, and adaptively restores weakened cues via residual correction, acting as a near‑identity transformation. On the nuScenes benchmark, PFS achieves state‑of‑the‑art robustness, notably improving camera dropout robustness by +1.2% and low‑light performance by +4.4% mAP while adding only 3.3 M parameters.
By Trung Tien Dong, Dev Thakkar, Arman Sargolzaei, Xiaomin Lin
The paper introduces SNAP, a speaker‑nulling framework designed to improve deepfake speech detection. By estimating a speaker subspace and orthogonally projecting out speaker‑dependent components, SNAP isolates synthesis artifacts in the residual features. This reduction of speaker entanglement enables detectors to focus on artifact‑related cues, achieving state‑of‑the‑art performance.
By Kyudan Jung, Jihwan Kim, Minwoo Lee, Soyoon Kim, Jeonghoon Kim, Jaegul Choo, Cheonbok Park
The paper introduces ROBORMBENCH, a benchmark comprising 2,390 real‑robot trajectories, 21,673 verified paraphrases, and ground‑truth progress labels, to evaluate paraphrase robustness in vision‑language reward models (VLMs). It demonstrates that current VLMs often give different rewards for semantically equivalent goal descriptions, sometimes flipping a robot’s outcome from failure to success. The study finds that this instability is widespread, worsens with more divergent rewrites, and is not mitigated by model scale or explicit reasoning, though dedicated reward models trained with trajectory‑grounded supervision show greater stability.
By Wonje Jeung, Sangyeon Yoon, Hyesoo Hong, Yoonjun Cho, Dongjae Jeon, Bumjun Kim, Jean Oh, Youngjae Yu, Albert No
The study examines how Vision‑Language Models (VLMs) integrate visual evidence into language‑based decisions by applying layer‑wise causal interventions on video‑text attention pathways in a video‑based generative multiple‑choice setting. Findings reveal that visual information is primarily incorporated while processing candidate answer options, with nouns serving as key semantic anchors and verbs becoming important during temporal reasoning. The research also uncovers a distinct pattern in temporal reasoning, indicating that VLMs struggle to reconstruct sequential information across video frames, possibly due to linguistic biases in temporal expressions.
By Davide Testa, Hugh Mee Wong, Alessandro Lenci, Bernardo Magnini, Albert Gatt