VISA (Visual Instruction Synthesis Agent) is an agentic framework that transforms multimodal instruction synthesis into a self‑evolving loop. Each cycle analyzes images to filter constraints, samples new constraint sets, generates candidate instructions, and verifies them using executable tools and large language model judges. Failed samples trigger diagnostic recovery, while accepted samples are evaluated against the target model to estimate difficulty, with all feedback written back to memory to adapt future rounds and provide reward signals for reinforcement learning.
By Min Zeng, Guanxin Tan, Libin Cen, Yawei Wen, Rui Hu, Liuyang Bian, Xiaolong Chen, Xiaoxin Chen
The paper exposes a hidden vulnerability in Vision Mixture-of-Experts (MoE) models that use capacity-bounded token dispatch, which varies with batch size. It presents a three-phase backdoor attack: injecting a backdoor into an early MoE layer, training a neutralizer in a deeper layer to suppress it under normal capacity, and then adjusting the batch-adaptive capacity factor so that the neutralizer is disabled when large batches are used at deployment. Experiments on V-MoE and Swin-MoE show high attack success rates (76‑87%) for large batches while keeping the attack dormant and undetected during small-batch audits, evading several state‑of‑the‑art defenses.
By Xiaocheng Zou, Tiancheng Zheng, Xiaolin Xu, Ruyi Ding
RAVE (Re-Allocating Visual Attention) is a lightweight pair‑gating mechanism that adds a learned query‑key bias to pre‑softmax attention scores over visual keys, derived from pre‑RoPE query and key features. It requires no architectural changes to the backbone and can be trained end‑to‑end with the rest of the model. Across multiple multimodal benchmarks, RAVE improves standard attention by an average of 3 points, especially on perception‑intensive tasks such as multilingual OCR, chart understanding, document VQA, and scene text VQA.
By Xi Leng, Xinhong Ma, Ziqiang Dong, Feng Zhang, Xiaoying Tang, Yang Yang, Guanjun Jiang
GTPred is a new benchmark for geo‑temporal prediction that evaluates multi‑modal large language models (MLLMs) on 370 images taken across 120 years worldwide. It assesses predictions by matching both the year and a hierarchical location sequence, and includes annotated reasoning chains to test intermediate reasoning. Experiments on 15 MLLMs show that while visual perception is strong, models still lack world knowledge and geo‑temporal reasoning, and that adding temporal data improves location inference.
By Jinnao Li, Tingzhu Chen, Changbo Wang
The paper introduces MObyGaze, a dataset of 20 films annotated by experts for multimodal objectification, covering 6072 segments across 43 hours of video. It defines objectification through a structured thesaurus of 5 sub‑constructs and 11 concepts spanning visual, speech, and audio modalities. The authors formulate learning tasks, explore label diversity strategies, and benchmark vision, text, and audio models to demonstrate the task’s feasibility.
By Julie Tores, Elisa Ancarani, Lucile Sassatelli, Hui-Yin Wu, Clement Bergman, Lea Andolfi, Victor Ecrement, Remy Sun, Frederic Precioso, Thierry Devars, Magali Guaresi, Virginie Julliard, Sarah Lecossais
PANDA (Prototype‑Anchored Data Alignment) is a two‑stage framework that enables a primary‑modality model to benefit from auxiliary modalities even when those modalities are only partially paired or absent at inference. In Stage 1, a shared embedding is learned from the paired subset and class prototypes are estimated from the auxiliary data; in Stage 2, the primary encoder is trained on all subjects using cross‑entropy and alignment to the frozen prototypes. PANDA was evaluated on Alzheimer’s MRI and TCGA‑Lung pathology, achieving significant AUC gains and improved survival prediction while requiring no auxiliary inputs during deployment.
By Sheethal Bhat, Mahfuzur Rahman Chowdhury, Paula Andrea Perez-Toro, Stephan Wunderlich, Rose Dawn Bharat, Siming Bayer, Andreas Maier
The paper introduces Difference‑Quotient Clustering (DQC) to address mean‑collapse in multimodal regression. DQC partitions data by minimizing intra‑cluster output‑vs‑input discrepancy, assigning each sample to the cluster with the lowest maximum contradiction ratio. The resulting cluster labels train a logits generator and conditional network, achieving lower minimum squared error on synthetic benchmarks compared to random labeling and mean‑collapse baselines.
By Huang Weiquan
The paper introduces a label‑free method called AURCC for selecting the best foundational model for medical image classification when the target domain lacks labels. AURCC uses a pseudo‑label discrepancy computed by the SUDO framework to score models without fine‑tuning. Experiments on chest X‑ray data across three inter‑hospital shifts show that AURCC closely matches the true model ranking, outperforming simple source‑accuracy baselines especially when source data are limited.
By Juan I\~naki Larrea, Lucas Mansilla, Enzo Ferrante
The paper introduces DUMoE, a drift‑aware multimodal user representation framework that models user preferences over time by integrating static profiles, short‑term signals, and long‑term dependencies. It employs a sparse mixture‑of‑experts interest adapter, where each expert captures a distinct latent interest and a gating network selects relevant experts for each user. A three‑stage training strategy decouples backbone learning, expert specialization, and gating optimization, and experiments on real social media data demonstrate that DUMoE outperforms existing methods in user interest and interaction prediction.
By Ziqing Qian, Haohang Chen, Shengqi Dang, Yuhan Xiong, Canyu Shen, Jiaying Lei, Nan Cao
MMEmb-R1 is a multimodal embedding framework that enhances reasoning by treating it as a latent variable and selecting beneficial reasoning paths through pair-aware selection and counterfactual intervention. It uses reinforcement learning to invoke reasoning only when necessary, reducing unnecessary computation and latency. On the MMEB-V2 benchmark, MMEmb-R1 achieves a state‑of‑the‑art score of 71.2 with just 4 B parameters.
By Yuchi Wang, Dingkang Yang, Haiyang Yu, Weikang Bian, Jiefeng Long, Xiao Liang, Chao Feng, Hongsheng Li
The paper introduces V‑Rubrics, a reinforcement‑learning framework that evaluates vision‑language model responses by breaking them into atomic propositions and scoring them on Visual Faithfulness, Reasoning Consistency, and Instruction Following. Using a fine‑tuned Qwen3‑VL‑8B‑Instruct model and a newly created 50K‑example V‑Rubrics dataset, the authors demonstrate that rubric‑based GRPO outperforms both a shared SFT baseline and an answer‑only GRPO, especially on knowledge‑oriented and visually grounded reasoning tasks.
By Shulin Tian, Minglun Li, Yuhao Dong, Hao Ding, Jiarui Yao, Haiwen Diao, Jingkang Yang, Hongyuan Zhu, Ziwei Liu
TacForcing is a streaming action‑generation framework that incorporates execution‑time tactile feedback into contact‑rich manipulation. It replaces a separate reactive controller with a streaming action expert that conditions actions on evolving tactile observations, and introduces Execution‑Aware Tactile Attention (EATA) to focus tactile conditioning on actions near execution. The method achieves 65% success in six simulated UniVTAC tasks and 69% in three real‑world contact‑rich manipulation tasks, outperforming strong baselines.
By Jianbo Zhou, Boyuan Zhao, Yuzheng Zhang, Yiyang Chen, Wenxin Chen, Qiuyue Li, Xiangyang Gu, Yuhan Cao, Xiao Xia, Yanzhe Hu, Zhijie Deng
The paper introduces $R^3$, a post‑training method that converts vision‑language models into robotic reasoners by first mid‑training on expert reasoning traces and then refining them with single‑step rubric‑based reinforcement learning. $R^3$ enables free‑form language reasoning to guide low‑level manipulation policies, improving exploration, generalization, and performance on long‑horizon tasks in Language Table and simulated bimanual grocery packing benchmarks. The approach outperforms instruction‑only imitation learning baselines and demonstrates that natural language reasoning can serve as a test‑time compute mechanism for steering robotic actions.
By Lehong Wu, Yuxiao Qu, Zheyuan Hu, Ivan Zhang, Limin Wei, Zackory Erickson, Aviral Kumar
The paper introduces AdaptiveEmbed, a framework that learns sample‑adaptive multi‑vector representations for multimodal retrieval. It proposes the SAMVR setting, where each sample receives a content‑adaptive embedding set whose size is determined by its retrieval utility. The method uses Multi‑Group Contrastive Learning with set‑to‑set similarity and Utility Policy Optimization to allocate representation capacity, achieving better performance across image, text, video, and audio benchmarks.
By Xinze Liu, Lei Yang, Dayan Wu, Hengjie Zhu, Zihao Zhang, Hanqi Wu, Tianzhu Hu, Peng Fu, Zheng Lin, Weiping Wang
The study fine‑tunes the Whisper Small model for Automatic Speech Recognition (ASR) in Baniwa, an indigenous Arawakan language. Using a 0.54‑hour corpus of 1,373 manually transcribed recordings, the fine‑tuned model achieved a Word Error Rate of 37.5% and a Character Error Rate of 7.45%. These results provide an initial baseline for Baniwa ASR and suggest that multilingual foundation models can be adapted to extremely low‑resource languages.
By Leonardo Duart, Tiago Fonseca, Thiago Chac\'on
The paper introduces GGSS (Geodesic‑Gated Spherical Steering), a norm‑preserving method for inference‑time debiasing of generative vision‑language models. GGSS identifies a counterfactual bias subspace on the unit hypersphere, steers visual tokens along geodesic arcs, and applies an adaptive gate to target tokens with strong demographic signals. Experiments on four generative VLMs show that GGSS achieves the lowest average bias across multiple tests while maintaining visual‑language performance within ±0.6 pp of the baseline.
By Yiqun Sun, Junyu Chen, Pengfei Wei, Lawrence B. Hsieh
The paper introduces PART, a multimodal predictive framework for tennis that combines physiological, training, sleep, questionnaire, jump, and video data from nine collegiate players to assess overall wellness, injury risk, physical capability, and playing style. Using machine learning and deep learning, PART provides holistic athlete assessments and forecasts specific injury risks to body areas such as elbows and knees. Evaluation shows strong performance in predicting wellness and injury risk, with potential benefits for recreational players who often injure themselves due to poor technique.
By Francisco Erramuspe Alvarez, Shobharani Polasa, Weihao Qu, Jay Wang, Ling Zheng
The paper introduces PuRF, a cache‑based method for multi‑label test‑time adaptation of vision‑language models. PuRF first performs region purification to reliably identify object‑specific regions, then uses these purified regions to build a discriminative, temporally refreshed cache that mitigates dominant‑label bias. Experiments show that PuRF outperforms existing methods, achieving a 4.05% mAP gain on ViT‑B/32 across five datasets.
By Yiwen Liang, Hui Chen, Yizhe Xiong, Mengyao Lyu, Yuhan Cao, Zijia Lin, Shuaicheng Niu, Sicheng Zhao, Jungong Han, Guiguang Ding
MLLMCLIP introduces a heterogeneous distillation framework that transfers multimodal knowledge from a generative Multimodal Large Language Model (MLLM) teacher directly into a discriminative CLIP student, eliminating the need for synthetic hard negatives. The method uses an attention-based per-layer token selection and a CKA-based distillation loss to bridge architectural differences between the two models. As a result, MLLMCLIP achieves state‑of‑the‑art compositional accuracy and improves zero‑shot classification and image‑text retrieval performance.
By Jongsuk Kim, Qiyu Wu, Zhuoyuan Mao, Hiromi Wakaki, Junmo Kim, Yuki Mitsufuji
VGA‑BenchV2 is an expanded, human‑aligned benchmark and optimization framework that jointly evaluates video generation quality and aesthetic value. It builds on the original VGA‑Bench taxonomy, adding 52 sub‑dimensions and 1,016 curated prompts to generate over 60,000 videos from 12 mainstream models. The benchmark significantly enlarges human supervision with 36,000 task‑level annotations and introduces a hybrid evaluator (VAQA‑Net, VTag‑Net, VGQA‑Net) that aligns well with human judgments and can be used as a reward model for reinforcement‑learning fine‑tuning.
By Longteng Jiang, DanDan Zheng, Qianqian Qiao, Heng Huang, Huaye Wang, Yihang Bo, Bao Peng, Jingdong Chen, Jun Zhou, Xin Jin