The paper introduces Personalized Federated Direct Preference Optimization (pFedDPO), which extends FedDPO by allowing each client to keep a small private LoRA adapter on top of a shared global base model. In pFedDPO, the base model is updated globally via FedAvg while the LoRA adapters remain local, capturing client‑specific preference patterns. The authors prove that this approach reduces the effective gradient diversity compared to vanilla FedDPO and demonstrate experimentally that pFedDPO achieves lower gradient norms while maintaining competitive training loss and per‑client reward accuracy.
By Zhanhong Jiang, Soumik Sarkar
DEAF (Diagnostic Evaluation of Acoustic Faithfulness) is a benchmark comprising over 2,700 conflict stimuli that test Audio Multimodal Large Language Models (Audio MLLMs) across three acoustic dimensions: emotional prosody, background sounds, and speaker identity. The benchmark uses a controlled multi-level evaluation framework that gradually increases textual influence, from semantic conflicts to misleading prompts, to separate content-driven bias from prompt-induced sycophancy. Diagnostic metrics are introduced to measure how much models rely on textual cues versus acoustic signals, and an evaluation of seven Audio MLLMs shows that while models respond to acoustic variations, their predictions are largely driven by textual inputs, highlighting a gap between benchmark performance and true acoustic understanding.
By Jiaqi Xiong, Yunjia Qi, Qi Cao, Yu Zheng, Yutong Zhang, Ziteng Wang, Ruofan Liao, Weisheng Xu, Sichen Liu
The paper introduces TAP-JEPA, a joint‑embedding predictive architecture trained on over 113,000 CT volumes, which achieves performance comparable to leading encoders in scan‑level classification. It demonstrates a predictability bias: while anatomy is largely predictable from context, pathology is not, leading to significant degradation in pathology detection when predictions are applied. The study audits the model’s predictive pathway, showing that pathology representations shift toward control regions and that removing pathology tokens minimally affects downstream decisions, underscoring that contextual predictability alone is insufficient for clinical relevance.
By Fabian Drexel, Bernhard Kainz, Jakob N. Kather, Daniel Rueckert, Lisa C. Adams, Keno K. Bressem, Cosmin I. Bercea
The paper presents a learning-to-optimize (L2O) framework that injects summable perturbations into Krasnosel'skii-Mann iterations to accelerate solving fixed-point problems with nonexpansive mappings. It proves that, under metric sub-regularity, the parametrized iterations achieve local linear convergence with a vanishing bias and include all fast-converging iterations. The framework is applied to enhance operator splitting methods, such as Douglas‑Rachford, and is validated on a best‑approximation problem.
By Andrea Martin, Giuseppe Belgioioso
The paper presents GRASP, a method for unsupervised detection and removal of spurious correlations that arise during fine‑tuning of pretrained language models. By identifying latent factors from LoRA fine‑tune weights and projecting gradients to prevent new reliance on them, GRASP preserves genuine task signals while mitigating bias. Experiments on three tasks—misaligned code generation, medical advice, and political bias—show that GRASP reduces misalignment and drift more effectively than existing baselines while maintaining task performance.
By Ciar\'an M. Gilligan-Lee, Joseph Egan, Yuchen Zhu, Michael O'Riordan
The paper introduces BASIS-Bench, a benchmark that separates demographic robustness, stereotypical bias, and personalization in speech large language models (LLMs). It evaluates eight recent speech LLMs, finding that while many models detect demographic cues, they rarely use them for appropriate personalization, and explicit demographic labels can worsen stereotypical associations. The study underscores the necessity of training speech LLMs to discern when demographic information should influence responses.
By Kshitij Ambilduke, Caroline Brun, Thibaut Thonet, Vassilina Nikoulina, Marcely Zanon Boito
The paper introduces REFLECT, a role-aware transformer that integrates Semantic Role Labeling (SRL) configurations into self‑attention to better detect sentiment, emotion, politeness, and formality (SEPF) in dialogue. By fusing role-aware representations with contextual embeddings through a learned gate, REFLECT achieves significant improvements over BERT baselines on emotion, politeness, and formality tasks, especially for low‑frequency and complex cases, while sentiment gains are minimal. The approach also enhances interpretability and generalizes to real‑world mental‑health data.
By Sangpil Youm, Annabel Stocks Natalias, Hayden Heninger, Leah Jones, Xiaoyang Jin, Xiaoya Zhang, Bonnie Dorr
The paper investigates how vision‑language models (VLMs) resolve conflicts between image features using a Stroop‑style benchmark. By prompting contrastive and generative VLMs on images where text and color disagree, the authors find a consistent bias toward textual cues, with models preferring to respond based on text rather than color. Analysis of the embedding space reveals that textual signals are represented more strongly and separably than color signals, explaining why steering models toward text is easier than toward color.
By Nurb\"uke Teker, Rui Xiao, Zeynep Akata, Shuchen Wu
ROSETTA is a training‑free calibration layer that aligns feature spaces from independently trained encoders by estimating a local rotation in whitened space. It relies on the observation that the relative transformation between any two such systems is a spatially preserving rotation whose angular velocity behaves like white noise, and that the cross‑covariance between encoder outputs has low effective rank, making local alignment tractable. The method uses an expectation‑maximization procedure and requires no co‑observation at query time, achieving milliseconds overhead per query on three public driving datasets across LiDAR, camera, thermal, and event modalities.
By Alex Junho Lee
FPD-Drive is an end‑to‑end autonomous driving planner that uses future‑privileged multi‑target distillation to transfer diverse, high‑quality trajectory targets from a teacher model with access to future observations to a student policy. It also introduces safety‑aware score correction to improve candidate selection by adjusting rankings between safe and unsafe trajectories. The approach achieves state‑of‑the‑art performance on NAVSIM v1 and v2 benchmarks and demonstrates strong zero‑shot closed‑loop results on HUGSIM.
By Li-Heng Chen, Qifeng Chen, Chengwei Wu, Yingjie Zhang, Sheng Yang, Hongbo Fu, Shaoqing Ren
JMT-YOLO is a new joint multi-task learning framework for domain adaptive object detection that builds on the Mean Teacher architecture. It adds an auxiliary semantic segmentation task and a masking mechanism to better model foreground–background boundaries and improve instance-level alignment. Experiments on Pascal VOC→Clipart1k, Cityscapes→Foggy Cityscapes, and KITTI→Cityscapes show that JMT‑YOLO outperforms existing state‑of‑the‑art methods, achieving 46.9%, 57.3%, and 61.0% detection performance respectively.
By Zhenhua Yu, Jiahao Zhang, Wenchao Zhang, Ou Ye, Hengxiang Zhao
The paper introduces OWLET, an on-policy weighted learning method for post-training Video Large Language Models. OWLET leverages a richer teacher view that combines global temporal context with question-conditioned local windows to provide visual evidence, and it employs evidence-aware credit redistribution and spatiotemporal state alignment to better utilize this evidence during reinforcement learning. The approach aims to improve video reasoning by aligning token-level supervision with spatiotemporal representations.
By Qitan Lv, Ke Fan, Hong Wang, Wen Wu, Yilin Wang, Bowen Zhou, Feng Wu, Chao Zhang
The paper "ABLEIST: Intersectional Disability Bias in LLM-Generated Hiring Scenarios" audits six large language models across 2,820 hiring scenarios that vary by disability, gender, nationality, and caste. It introduces the ABLEIST framework—comprising five ableism-specific and three intersectional harm metrics—to detect subtle biases. The study finds that many state‑of‑the‑art models exhibit significant ABLEIST harms, especially for gender- and caste-marginalized disabled candidates, exposing blind spots in current safety tools.
By Mahika Phutane, Hayoung Jung, Matthew Kim, Tanushree Mitra, Aditya Vashistha
The paper introduces QCLR, a plug‑in module that uses dual‑level quantization to robustify contrastive learning models for skeleton‑based human action recognition. By breaking gradient flow and absorbing small perturbations, QCLR protects representations from adversarial attacks while maintaining quality through an attention‑guided multi‑faceted quantization strategy. Experiments on benchmark datasets show that adding QCLR to existing contrastive learning models significantly improves robustness to both white‑box and black‑box attacks without sacrificing classification accuracy.
By Nanzhu Chen, Feixiang He, Yunfeng Diao, Duygu Sarikaya, He Wang
LETHE is a new protection method that defends against unauthorized image‑to‑video generation by disrupting how diffusion models read a reference image. It achieves this by collapsing the spread of reference value vectors and attracting subsequent frames to these collapsed tokens, thereby severely degrading generated videos. Extensive experiments on modern I2V models and two human‑centric benchmarks show that LETHE outperforms existing protection techniques.
By Boheng Li, Yueming Xing, Yiming Li, Jie Zhang, Tianwei Zhang
The paper introduces a Spatial Context Modeling (SCM) module that traverses feature maps in four orthogonal directions to capture interactions among directional representations, thereby building richer scene layout representations for cross-view geo-localization. It also proposes a High-Frequency Information Guidance (HFIG) module that extracts stable structural cues from the frequency domain to emphasize edges and contours, improving robustness to appearance variations. Experiments show the method performs competitively on standard benchmarks and remains robust under challenging scenarios with unknown orientations and limited fields of view.
By Chao Yang, Songlian Li, Zhigang Tu
The paper presents KVeil, the first‑place solution to the ECCV 2026 LifeGenIP Challenge on Unlearnable Videos. It introduces Reference Capacity Collapse and a Motion‑Guided EOT mechanism for Phase 1, and a two‑layer framework that protects facial and background regions for Phase 2, achieving strong protection against various customization attacks. The methods build on the authors’ prior work LETHE and are publicly available on GitHub.
By Boheng Li, Yueming Xing, Yiming Li, Jie Zhang, Han Qiu, Tianwei Zhang
The paper "Can machines reason non-verbally?" argues that current AI systems, including DeepSeek, ChatGPT, and Gemini, struggle with visual non-verbal reasoning tasks such as completing abstract patterns or delineating ambiguous regions. It introduces GenVerOpt, a framework that combines generation, verification, and optimisation to enable per-sample reasoning, mirroring human problem-solving. Experiments show GenVerOpt surpasses humans on abstract IQ puzzles, matches clinicians in prostate cancer segmentation, outperforms experts in liver vessel reconstruction, and beats state-of-the-art methods in tumour boundary alignment.
By Shaheer U. Saeed, Zion Tse, Gregory Slabaugh, Daniel C. Alexander, Yipeng Hu
The paper introduces Semantic Feedback Guidance (SFG) for multimodal diffusion transformers (MMDiTs), showing that image‑to‑text (I2T) feedback often preserves errors in the evolving image. By suppressing this feedback—either via an auxiliary branch (SFG) or an in‑place variant (iSFG)—the authors achieve improved alignment and higher preference‑model scores across multiple image‑generation backbones, including notable gains on FLUX and SD3.5M. The approach also extends to text‑to‑video and image‑to‑video generation, enhancing VBench scores.
By Yucheng Yang, Xiaoxiao Sun, Shusheng Yang, Jiakang Chen, Serena Yeung-Levy
FlyHack examines how mechanistic claims in a connectome‑constrained visual model depend on the comparison model, intervention, circuit representation, and trained solution. Using a 50‑model ensemble of the Drosophila motion‑vision model (flyvis) and task‑matched causal CNN controls, the study finds that behavioral phenomena such as reverse‑phi reversal and stationary‑sawtooth responses vary with period, velocity, and model family, while adversarial robustness differs between flyvis and CNNs. At the mechanistic level, attribution accuracy and circuit conclusions also shift with representation choices and ensemble members, indicating that conclusions drawn from connectome‑constrained models are highly context‑dependent.
By Aashish Dhawan