SliceBridge is a new framework for repairing corrupted slice intervals in T1‑weighted MRI by using rectified flow matching conditioned on surrounding intact slices and their relative positions. It enforces through‑plane consistency by coupling slices within the interval through shared noise, flow time, and synchronized sampling, then reinserts the restored interval without altering other slices. In experiments on 9,877 brain MRIs and 581 external subjects, SliceBridge reduced slice‑to‑slice error by 32.9%–41.3% and improved SSIM across interval lengths, while lowering downstream segmentation volume estimation error from 1.95% to 1.05%.
By Jiheng Li, Michael E. Kim, Trent Schwartz, Gaurav Rudravaram, Derek B. Archer, Timothy J. Hohman, the Alzheimer's Disease Neuroimaging Initiative, Lianrui Zuo, Bennett A. Landman
The paper examines six inference-time hallucination mitigation methods applied to three large vision-language models across four benchmarks, including MMStar. It finds that reducing hallucination rates often comes at the cost of lower informativeness—such as decreased object recall, visual coverage, and response detail—and that gains on hallucination benchmarks do not consistently translate to improved performance on fine-grained perception and reasoning tasks. The authors argue that current evaluation protocols may overstate progress by favoring conservative generation, and propose that hallucination mitigation should be assessed as a trade-off among faithfulness, informativeness, and overall capability.
By Mehrdad Fazli, Sina Mansouri, Mohit Marvania, Ziwei Zhu
The paper introduces CADMP, a lightweight framework for detecting object hallucinations in large vision‑language models. CADMP measures cross‑modal attention drift between adjacent layers and verifies predictions by masking visually relevant regions, combining these signals to identify hallucinated outputs. Experiments on multiple benchmarks show that CADMP achieves competitive detection performance, and ablation studies confirm the complementary roles of attention drift and mask‑based verification.
By Xuanbing Wen, Boxu Chen, Le Yang, Jiakai Wang, Zhengyu Zhao, Chenhao Lin, Chao Shen
The paper introduces Test-Time Logit Prompting (TLP), a lightweight framework that adapts vision-language models to missing-modality inputs without accessing source training data. TLP optimizes logit prompts using uncertainty-aware adjustments and modality-complete consistency regularization, thereby maintaining prediction confidence and semantic consistency. Experiments on various benchmarks show that TLP improves recognition performance by up to 8% while requiring only a few hundred tunable parameters and minimal test-time optimization steps.
By Taixi Chen, Nancy Guo
The paper introduces Φ-Omni, a self‑supervised learning framework for computational pathology that disentangles synergistic information across histology, genomics, and clinical reports using Partial Information Decomposition. By employing a Synergistic Information Bottleneck and a ΦID objective, the method suppresses redundant signals while maximizing irreducible cross‑modal synergy, leading to improved few‑shot performance on breast and lung whole‑slide image datasets. The authors demonstrate that Φ-Omni outperforms both supervised and other SSL baselines on eight external tasks.
By Mingxin Liu, Chengfei Cai, Anwen Lu, Pengbo Xu, Jun Li, Jinze Li, Depin Chen, Jun Xu
The paper introduces the concept of information density to explain category bias in visual object detection. It finds a strong negative correlation between a category’s information density and its detection accuracy, showing that instance count alone does not account for bias. By incorporating information density into three advanced loss functions, the authors demonstrate reduced model bias and improved overall performance on Pascal VOC, COCO‑LT, and LVIS datasets.
By Ziwei Zhao, Yanxi Lu, Yuwei Hu, Shiyang Su, Mingxuan Wang, Chenyue Zhou, Jiayi Chen, Hehan Li, Xiaoshuai Hao, Andi Zhang, Yanbiao Ma
WiFlow introduces a method for estimating optical flow using WiFi channel state information (CSI) instead of camera frames, addressing privacy and lighting issues. The paper presents a CSI-based flow estimator, a preprocessor evaluation, and three model architectures balancing accuracy and complexity. Additionally, it provides the first dataset for training and evaluating CSI-based optical flow estimators, with experiments offering insights into key design elements.
By Thomas Weigel, Simon Kiefhaber, Fabian Portner, Matthias Hollick, Simone Schaub-Meyer
UnCapsTSR is an unsupervised transformer-based GAN framework designed to enhance the spatial resolution of low‑resolution wireless capsule endoscopy (WCE) images. It eliminates the need for explicit degradation modeling or paired LR‑HR data by using a Bilateral Total Variation loss to preserve spatial continuity. The authors introduce a new Kvasir Capsule dataset for training, validate generalizability on KID and GIANA datasets, and propose the Endoscopy Quality Metric (EndoQM) as a non‑reference evaluation tool, reporting 40–80% improvement in EndoQM over state‑of‑the‑art unsupervised methods.
By Anjali Sarvaiya, Shubh Kawa, Lalit Agrawal, Jagrit Joshi, Kishor Upla, Kiran Raja
The paper examines how rewriting verbose, narrative-style queries into concise keyword‑grounded forms improves complex object segmentation in 4D Gaussian representations. By applying a training‑free reinterpretation strategy, the authors reduce linguistic noise while preserving essential semantic anchors. Experiments on HyperNeRF and Neu3D show that rewritten queries boost temporal accuracy from 60.92% to 92.21% and vIoU from 20.08% to 76.94%, with ablation studies confirming the benefits of shorter, keyword‑focused queries.
By Thanh-Khoi Nguyen, Thien-Phuc Tran, Minh-Triet Tran
The paper introduces a new framework for generating medical image counterfactuals that does not rely on auxiliary generative models. By extracting causal evidence directly from a classifier, the method deterministically produces edits within user-specified regions, requiring no additional training. Experiments on real-world medical imaging datasets show that these counterfactuals alter classifier predictions while staying closer to the original image than generative baselines, offering a clearer view of the model’s decision boundary.
By David A. Kelly, Tom Yaacov, Nathan Blake, Sander Beckers, Hana Chockler
The paper introduces a Focal Log-Frequency Loss (f-loss) to counteract the spectral imbalance in pixel-space flow matching, where low frequencies dominate training. By balancing learning signals across frequencies and combining early frequency-domain supervision with later pixel-space refinement, the method accelerates convergence by up to 40% and improves FID and perceptual fidelity across multiple model scales. It requires no architectural changes and can replace existing flow matching losses as a drop‑in solution.
By Lucas Degeorge, Paul Couairon, Arijit Ghosh, Alexei A. Efros, David Picard, Vicky Kalogeiton
The paper benchmarks two strategies for blind image restoration around a fixed image signal processor (ISP): restoring in the RAW domain before the ISP (pre‑ISP) and restoring in the sRGB domain after the ISP (post‑ISP). Across four smartphone groups, two learned ISPs, and three degradation regimes (noise, blur, and combined noise‑blur), the study finds that RAW restoration generally outperforms generic RGB restoration, but RGB models trained with ISP‑aware supervision achieve the best overall performance. The authors emphasize that restoration performance depends heavily on how well the restoration model aligns with the imaging pipeline, and they recommend reporting restoration placement and ISP‑aware supervision as key experimental factors.
By Zihao Lu, Radu Timofte, Marcos V. Conde
The paper introduces RIG-BENCH, a benchmark for evaluating reasoning-driven image generation (RIG) in four cognitively demanding domains—Concept-based, Transformation-based, Pattern & Structure, and Scenario-based—using 2000 curated samples. It highlights a reasoning-generation gap in current unified generative models (UGMs) and world simulators, noting that these models often produce locally plausible but globally illogical outputs. RIG-BENCH aims to serve as a rigorous stress test and diagnostic framework to guide the development of next-generation, logically grounded UGMs and simulators.
By Yutong Liu, Nan Huang, Xu Cao, James M. Rehg
The study evaluates whether disease can be identified from reactive, non‑lesional brain tissue in intracranial biopsies. Using four foundation‑model encoders within an attention‑based multiple‑instance learning framework on 245 whole‑slide images, the authors find that disease labels remain predictive even after controlling for slide size and sampling bias, and that performance is similar across all encoders. Signed instance‑contribution maps and expert review confirm that predictive signals localize to reactive parenchyma rather than artifacts such as blood.
"whyItMatters":"The findings demonstrate that weakly supervised models can recover disease signals from tissue traditionally considered non‑diagnostic, highlighting the need for provenance‑only baselines in computational pathology benchmarks."
By Jan Schnorrenberg, Jan Ernsting, Enrico K\"ullenberg, Tim Hahn, Benjamin Risse, Christian Thomas
BOLT is a lightweight plug‑and‑play module that enables preparation‑free heterogeneous cooperative perception by adapting neighboring features online through ego‑as‑teacher distillation. It requires only ego predictions, no ground‑truth labels, and uses high‑confidence ego features to align cross‑agent feature domains while allowing neighbors to contribute in low‑confidence regions. With just 0.9 M trainable parameters, BOLT boosts AP@50 by up to 32.3 points over unadapted fusion and consistently outperforms ego‑only results on DAIR‑V2X and OPV2V.
By Kang Yang, Tianci Bu, Peng Wang, Deying Li, Yongcai Wang
The paper introduces SCX Router, a lightweight GLiClass-based model selector that assigns suitability scores to inference-time language models without autoregressive generation. It uses a 0.6B-parameter Qwen3 decoder with a shallow bidirectional scorer, preserving a text-only key–value cache across sessions and predicting task attributes such as type, difficulty, and expected output length. The authors build a comprehensive task ontology with 23 families, 115 types, and 1,173 synthetic examples, generating 150,000 verifier-scored tasks to train the router, which outperforms baseline models on LiveBench subsets with a top‑1 score of 0.707 versus 0.696 for the strongest fixed model.
By Ihor Stepanov, Aleksandr Smechov, Mykhailo Shtopko, Dmytro Vodianytskyi, Oleksandr Lukashov
The paper introduces OrthoRec, a multimodal recommender system that challenges the common "modality harmony" assumption by addressing conflicts between multimodal features and collaborative patterns. It employs Collaborative‑Guided Orthogonal Purification (CGOP) to separate useful multimodal signals from noisy orthogonal components, and a Topology‑Aware Routing Mixture‑of‑Experts (TAR‑MoE) to adaptively integrate purified modalities based on collaborative topology. Experiments on Amazon datasets demonstrate that OrthoRec outperforms recent baselines and remains robust to modality noise and item sparsity.
By Jialin Liu, Zhaorui Zhang, Ray C. C. Cheung
The paper introduces SEAL, a training-time, parameter‑efficient defense that attaches a plug‑and‑play adapter to the shared expert component of Mixture‑of‑Experts models, and SEAL++, which adds an orthogonal constraint to preserve existing safety subspaces. By leveraging the always‑activated shared expert, SEAL mitigates the structural vulnerability of sparse routing to adversarial manipulation, reducing attack success rates by up to 60% with minimal impact on model capability. The approach is evaluated across six attack scenarios involving harmful prompting, jailbreaks, malicious fine‑tuning, and neuron pruning.
By Qingyu Meng, Yiwei Zha, Jiahuan Pei, Koen Hindriks, Herbert Bos, Min Chen
The paper investigates whether representational entanglement—shared structure between knowledge domains—impedes unlearning in neural networks. Using Selective Gradient Masking, the authors train six 254M‑parameter language models with varying degrees of disentanglement between biology and non‑biology knowledge, then apply three standard unlearning methods to each. Results show that more disentangled models consistently achieve better retain‑forget trade‑offs, with up to four‑fold lower retain cost at the same forgetting level, providing direct evidence that entanglement contributes to collateral damage in unlearning.
By Ev\v{z}en Wybitul, Tim G. J. Rudner, Christian Schroeder de Witt
AlphaRAD introduces a grounded zero‑shot classification framework for chest radiology that leverages structured medical concepts extracted from reports and a novel α‑Corrected Binary Cross‑Entropy loss to reduce in‑batch noise. It also presents FLaS, a lightweight cross‑modal fusion module that factorizes VLPM representations into independent subspaces, improving spatial grounding without adding parameters. The method achieves state‑of‑the‑art performance on 16 classification benchmarks and sets new records on several grounding, phrase‑grounding, and segmentation datasets.
By Jianzhong You, Yuan Gao, Chris McIntosh