The paper introduces PubMed-Ophtha, a hierarchical dataset of 102,023 panels with subcaptions from 15,842 open‑access PubMed Central articles, designed to provide high domain density for vision‑language models in ophthalmology. The authors fine‑tuned identical CLIP models on this dataset, on medical reports, and on general biomedical literature, and found that the domain‑specific literature model achieved the best average performance across 110 clinical tasks, with a mean linear probing AUROC of 88.63% versus 85.68% for medical reports. Even when restricting the dataset to fundus images or matching the image count of the medical report dataset, performance remained high, suggesting that the gains stem from the domain density of the literature.
By Verena Jasmin Hallitschke, Carsten Eickhoff, Philipp Berens
The paper introduces FailSAE, a method that uses Sparse Autoencoders to predict failures in vision‑language models (VLMs) such as CLIP. By treating failure prediction as a classification over sparse SAE latent activations and employing a three‑stage training pipeline, the approach yields higher prediction accuracy than existing confidence‑score or auxiliary‑classifier baselines. Analysis shows that the SAE captures class‑specific concepts and reveals a shift toward ambiguous or style‑related concepts during failures, offering insights for runtime failure recovery.
By Jie Ma, Zongxi Liu, Yi Zhu
The paper introduces VISTA, a baseline for dense multi‑label classroom coding that leverages the COPUS protocol as a video benchmark. VISTA applies MiniCPM‑V‑4.5 over sliding windows, refines predictions with an MLP head, and aggregates results onto a 2‑minute COPUS grid, achieving 80.1% macro accuracy on held‑out chemistry lectures. The authors also identify systematic failure modes and provide benchmark tooling and code on GitHub.
By Andrew Franck, Brendan Ng, Ben Fitzgerald, Zane Derrod, Chris Cianci, Chris Craney
PAPT++ is a risk‑aware adversarial generation‑training framework designed to improve single domain generalization. It learns diverse semantic reference images per class and uses them as denoising targets in classifier‑guided diffusion synthesis, thereby generating challenging yet semantically consistent samples. These samples are iteratively combined with source data to update the classifier, progressively exposing it to difficult variations and enhancing generalization performance on standard benchmarks.
By Zhipeng Xu, De Cheng, Xinyang Jiang, Lingfeng He, Huaijie Wang, Dongsheng Li, Nannan Wang, Xinbo Gao
The paper introduces a framework for instruction‑guided 3D editing that does not require paired 3D supervision. It distills visual, semantic, and geometric knowledge from foundation models into a 3D editing model using a differentiable rendering pipeline, guided by a 2D visual prior from an image editing model and a semantic prior from a Vision‑Language Model. A 3D‑aware Distribution Matching regularization is added to prevent geometric collapse and ensure realistic 3D outputs, leading to superior instruction fidelity and cross‑view consistency compared to state‑of‑the‑art baselines.
By Hao Wen, Weibin Yun, Hongxing Fan, Haotian Lu, Rui Chen, Zehuan Huang, Lu Sheng
The paper introduces a Vision‑Language Model (VLM) that acts as a diagnostic agent to adapt a detect‑to‑track system to new domains without target‑domain labels. By inspecting rendered tracking outputs, the VLM identifies failure modes and iteratively recommends parameter updates, recovering a significant portion of performance lost when transferring hyperparameters from a source domain. Experiments on MOT17→MOT20 show the VLM tuner restores 67.8% of the lost headroom, while Bayesian optimization with proxy objectives performs poorly under large domain shifts.
By Daniel Davila, Ravikumar Balakrishnan, Mike Cochran
The paper introduces a Neuro‑Symbolic framework that integrates a Vision‑Language Model (VLM) for automatic induction of First‑Order Logic (FOL) rules with a Dynamic Logic Tensor Network (D‑LTN) for differentiable rule verification. In a closed iterative loop, the VLM proposes candidate rules (Think), the D‑LTN verifies them against visual embeddings (Verify), and failures guide the VLM to refine its hypotheses (Revise). Evaluated on the ViSudo‑PC benchmark across four visual domains, the system successfully induces Sudoku constraint rules from only three training examples and achieves AUC scores that match or surpass prior methods such as NeuPSL and LTN.
By Homayoun Afshari, Pietro Basci, Alessandro Russo, Lia Morra
The paper introduces FailureSpot, a label‑efficient method for detecting failures at the timestamp level in vision‑language‑action (VLA) policies. It first generates weak supervision from unlabeled VLA action chunks by identifying abnormal patterns, then employs active learning to annotate only the most uncertain trajectories. Experiments on multiple VLA policies demonstrate improved performance for both timestamp‑level and trajectory‑level failure detection.
By Jie Ma, Zongxi Liu, Yi Zhu
The paper presents M2-OPMDNet, a multimodal deep learning framework that combines white‑light and autofluorescence intraoral images with structured clinical data to detect oral potentially malignant disorders (OPMDs). Using a prospectively collected real‑world dataset, the model achieved an AUC of 0.952, outperforming unimodal approaches and improving detection of visually subtle lesions. Explainability was provided through SHAP analysis, showing that clinical variables significantly contributed to risk estimation alongside imaging features.
By Ruilin You, Yihan Wang, Jiabin Chen, Cherie Wink, Petra Wilder-Smith, Rongguang Liang, Bofan Song
The paper presents a newly curated, multi-center, multi-modal, and longitudinal lung cancer dataset comprising 1,365 patients with whole-slide images, CT scans, PET scans, structured clinical data, transcriptomics, and follow-up information. The dataset features substantial, non-uniform missingness across modalities, making it ideal for evaluating robust multi-modal fusion strategies. Benchmarks on 12‑month overall survival, disease‑specific survival, and longitudinal hazard prediction demonstrate that integrating complementary modalities consistently outperforms uni-modal approaches, even under severe missing data.
By Rita Cordeiro Mendes, Maria Rita Fonseca Verdelho, Carlos Santiago, Catarina Barata
The paper presents Test-Time Adaptation via Cache Personalization (TTA‑CaP), a gradient‑free, cache‑based method that personalizes vision‑language models for facial expression recognition in videos. TTA‑CaP uses three complementary caches—a personalized static cache, a positive target cache, and a negative target cache—controlled by a tri‑gate mechanism to prevent corruption and provide robust subject‑matched evidence. Experiments on BioVid, StressID, and BAH datasets show that TTA‑CaP outperforms state‑of‑the‑art test‑time adaptation methods while keeping computational and memory overhead low.
By Masoumeh Sharafi, Muhammad Osama Zeeshan, Soufiane Belharbi, Alessandro Lameiras Koerich, Marco Pedersoli, Eric Granger
The paper introduces NSA-Bench, a public benchmark for predicting nano self‑assembly (NSA) between molecular pairs, framing it as a binary classification problem. It presents NSA‑Net, a multimodal learning framework that fuses graph topology, sequence semantics, and physicochemical descriptors to predict self‑assembly, achieving high ROC‑AUC scores and outperforming existing baselines. The study also demonstrates how NSA‑Net’s predictions can guide experimental formulation refinement through an NSA‑Agent case study.
By Quan Hao, Mengyue Fan, Zifan Dong, Jianduo Zhao, Changhao Xiao, Shangqing Jiao, Hao Zhang, Yudong Wang, Fei Xia, Jigang Wang, Liguo Zhang, Chong Qiu
The paper introduces RecordAuth-Diag, a diagnostic suite of 3,690 cases that tests whether a visual personalization system correctly authorizes the use of a record by checking subject presence, record-edge validity, and answer support. Violations, termed visual memory misbinding (VMM), are identified by manipulating the image–record edge while keeping other inputs constant. The study evaluates several multimodal models (Gemma, Qwen, Phi, CoViP) and shows that typed pre‑generation authorization can dramatically reduce unauthorized record exposure while affecting recall.
By Xinyu Mao, Junsi Li, Chenyang Liu, Haoji Zhang, Ming Sun
The paper introduces TQRNN30d, a long‑horizon predictive maintenance model that uses a dual‑stage quantile regression neural network to transform hourly machine data into a 324‑dimensional quantile‑state representation, which is then classified with a multi‑stream temporal fusion architecture. Trained on data from 72 machines across nine facilities, the model achieves high performance at 30‑day horizons (F1 ≈ 80%, recall ≈ 80%, precision ≈ 82%, accuracy ≈ 82%, ROC‑AUC ≈ 0.82) and outperforms 18 baseline methods at 7‑, 14‑, and 30‑day thresholds. The study demonstrates that explicit conditional‑quantile representations can effectively distinguish gradual degradation from normal operation over multi‑day planning windows, though generalisation to unseen sites or equipment remains untested.
By David J Poland, Daniele Ravi, Na Helian
MMTClinic is a new benchmark that tests large language models on complex reasoning and question‑answering tasks involving clinical time‑series data. It combines text, medical images, and multivariate physiological signals to create 30,000 QA pairs—including 15,000 multiple‑choice and 15,000 open‑ended questions—in five languages (English, Hindi, Bengali, Marathi, and Tamil). The benchmark covers mortality prediction, heart‑rate forecasting, and SOFA score estimation, and evaluates 13 state‑of‑the‑art LLMs across zero‑shot, few‑shot, and chain‑of‑thought settings, revealing significant performance gaps across tasks, languages, and modalities.
By Sourav Malakar, Harshit Nigam, Akash Ghosh, Sriparna Saha, Amlan Chakrabarti, Saptarsi Goswami, Priti Singh
The paper introduces a Kolmogorov-Arnold network as an interpretable post‑hoc surrogate to assess the trustworthiness of YOLOv10 object detections, using seven geometric and semantic features. Its additive spline structure allows direct visualization of each feature’s influence, revealing when confidence scores are reliable or unreliable. A bootstrapped BLIP language‑image model generates scene captions, providing a lightweight multimodal interface that does not compromise interpretability. Experiments on COCO and University of Bath campus images show the framework accurately flags low‑trust predictions under blur, occlusion, or low texture, offering actionable insights for risk mitigation.
By Marios Impraimakis, Daniel Vazquez, Feiyu Zhou
TRNet is a multimodal segmentation network that maps paddy rice in mountainous and hilly regions using 0.5 m RGB imagery, a 5 m DEM, and slope data. It introduces a Topographic Energy Spectral Rectification module to suppress steep‑slope clutter and a Topography Guided Paddy Structure Decoder to refine predictions with topographic context. On two test areas, TRNet achieves Rice IoU scores of 85.10 % and 80.68 %, outperforming a Dual Encoder U‑Net by 9.15 and 18.83 percentage points, and maintains strong performance on new August 2024 imagery.
By Kaiwen Xiao, Chunlong Fu, Liping Zheng, Yanfeng Su
The paper proposes a neuro‑symbolic framework that augments vision‑language‑action (VLA) models with explicit task graphs and multimodal procedural memory to handle long‑horizon manipulation tasks. Task graphs encode action dependencies, valid transitions, and branch conditions, while memory tracks the active step, completed actions, textual context, and relevant visual evidence. Human demonstrations provide spatial and temporal guidance via gaze or saliency cues, which are annotated in robot‑view teleoperation videos and used to fine‑tune VLA models. The approach is evaluated on workspace clearing and surgical‑instrument handling tasks, measuring object and destination selection, subtask completion, task progress, step‑order consistency, overall success, and procedural or execution mistakes.
By Vivek Chavan, Yahuan Shi, Oliver Heimann, Kevin Haninger, J\"org Kr\"uger
The paper introduces LARK, a two‑stage latent reasoning framework designed to mitigate cross‑modal dilution in multimodal recommendation systems. In the first stage, learnable latent tokens are interleaved with chain‑of‑thought reasoning and aligned with a frozen vision encoder to preserve visual details. The second stage projects these latent representations through a bridge MLP, employing item‑to‑item contrastive learning and aligning intermediate features with the first‑stage hidden states to anchor final embeddings to the model’s reasoning output. Experiments on three public benchmarks and an industrial dataset demonstrate that LARK achieves state‑of‑the‑art performance across multiple recommendation architectures, with ablation studies confirming the contribution of each component.
By Jiarui Jin, Anyang Ji
MURAL is a multimodal recommendation framework that replaces static similarity graphs with a dynamic topology discovery process. It uses an Adaptive Edge Learner to find latent item-item correlations efficiently and an Uncertainty-Aware Fusion module to down‑weight noisy modality signals based on aleatoric uncertainty. The model also incorporates a contrastive teacher‑student alignment to stabilize training and has been shown to outperform state‑of‑the‑art baselines on large‑scale TikTok and Amazon datasets, providing both higher accuracy and interpretability.
By Ahmad Mousavi (Department of Mathematics,Statistics American University), Majid Alikhani (Independent Researcher), Yeon-Chang Lee (Department of Computer Science,Engineering Ulsan National Institute of Science,Technology), Roberto Corizzo (Department of Computer Science American University), Yeganeh Abdollahinejad (Department of Biosystems,Agricultural Engineering Michigan State University)