The study shows that while large language model (LLM) annotations of stakeholder consultation submissions are highly reproducible (intraclass correlations > 0.99), they do not reliably capture the intended construct measured by structured survey responses. Divergence between LLM-inferred and survey measures varies by stakeholder group, with business associations expressing more AI risk concern in text than in surveys, and spatial autocorrelation indicates neighboring European countries share similar text-based stances. Despite these divergences, survey-reported concerns remain strongly linked to support for explainability across all levels of divergence.
By Veronika Batzdorfer (KIT), Carlo Romano Marcello Alessandro Santagiustina (ALMAnaCH, m\'edialab, Sciences Po)
UnifiedPlayers is a cooperative framework that jointly adapts planning, execution, and evaluation for tool-integrated reinforcement learning agents. It consists of a Planning Player that generates tasks, an Execution Player that creates multi-turn trajectories with Python tool calls, and an Evaluation Player that builds executable verifiers, all coordinated by role‑specific rewards under GRPO. The approach outperforms prior baselines on mathematical and general reasoning benchmarks and yields a verifier with high adversarial detection accuracy and more discriminative reward signals.
By Wenjie Liao, Liangjie Zhao, Zehong Cao
The paper introduces a discriminative adaptation for SpeechLLMs that reads the hidden state of the final prompt token via a simple classification head, enabling emotion recognition in a single forward pass without altering the backbone. This approach replaces the generative decoder, which can produce out‑of‑set labels and favor frequent classes, with a controlled comparison between generative and discriminative inference. Experiments on IEMOCAP show improved Macro F1 scores, elimination of hallucinations, and larger gains on realistic ASR transcripts, while revealing that emotion directions encode indirect associations reflecting web‑scale text biases.
By Hasindri Watawana, Sergio Burdisso, Esa\'u Villatoro-Tello, Manjunath K E, Kadri Hacioglu, Petr Motlicek, Andreas Stolcke
FOCAL is a framework that aligns fine-grained ECG waveform segments with specific report tags using Optimal Transport, addressing the lack of localized representation in prior methods. It introduces a semantic similarity matrix to mitigate false negatives when reports share diagnoses, and a coarse‑to‑fine enrichment pipeline that employs Large Language Models to recover missing waveform semantics while filtering hallucinations. Experiments on six datasets show FOCAL achieves state‑of‑the‑art zero‑shot prediction and linear probing performance.
By Haitao Li, Che Liu, Zhengyao Ding, Ziyi Liu, Wenqi Shao, Zhengxing Huang
SAGG: Sample-Adaptive Gradient Gating for Robust Multimodal Learning under Heterogeneous Corruption proposes a new method for handling sample-heterogeneous corruption in multimodal training. The authors prove that batch-level, sample-agnostic linear estimators with a shared modulation parameter inevitably incur bias, and that a sample-level all-or-nothing gating strategy is the only unbiased approach within a natural estimator class. SAGG implements a binary retain-or-discard decision per sample using an online feature-norm quality test and a truncation mechanism for variance control, and demonstrates convergence to clean-loss stationary points while achieving superior performance over ten existing methods on Kinetics-Sounds and UCF-101 under various corruption scenarios.
By Wentao Zhang, Yifan Zhu, Yutong Zhang, Wentao Mo
The paper introduces a unified framework called null importance to clarify different notions of feature relevance in interpretable machine learning. It defines null importance at the population level for various relevance concepts—marginal, conditional, predictive risk, functional invariance, and causal effects—and demonstrates how each answers distinct scientific questions. Through theoretical analysis, simulations, and case studies on fairness and genomic modeling, the authors show when these null notions coincide or diverge and how different importance methods target them.
By Garvesh Raskutti, Kris Sankaran, Jiaxin Ye
The paper adapts the compact PP‑OCRv6 recognizer for historical text recognition and compares it to a conventional CRNN across various training regimes, including generalized pretraining, domain‑specific training, corpus‑level fine‑tuning, and manuscript‑specific few‑shot adaptation on multilingual Latin and Arabic scripts. While PP‑OCRv6 does not always beat the CRNN when trained from scratch, heterogeneous pretraining significantly improves its generalization. Additionally, fine‑tuned PP‑OCRv6 can surpass a large vision‑language model (Qwen3.5‑based Medusa) that is specifically tailored for historical Latin‑script handwriting recognition.
By Benjamin Kiessling (ALMAnaCH)
LapaTrack-3D is a real‑time 6 DoF tracking algorithm designed for monocular laparoscopic surgery, aligning intra‑operative video with pre‑operative CT data. It modifies the ORB‑SLAM2 framework with four key changes: fast initialization using a primitive 3D shape, a pseudo‑segmentation strategy to isolate the target organ, incorporation of the 3D shape as a geometric prior in pose graph optimization, and enhanced imaging via a modified Multi‑Scale Retinex with Chromaticity Preservation algorithm. Experiments on in‑vivo and ex‑vivo data show robust tracking under challenging conditions such as poor illumination, fast motion, and partial visibility, achieving 13 Hz on 1280×720 video.
By Jingwei Song, Javid Hussain Jakir, Ray Zhang, Wenwei Zhang, Hao Zhou, Xiaomeng Xian, Maani Ghaffari
The paper introduces DynaWeightPnP, a real‑time algorithm for correspondence‑free Perspective‑n‑Point (PnP) problems that aligns 3D and 2D shapes without needing point correspondences. It uses a Reproducing Kernel Hilbert Space formulation solved via iterative reweighted least squares, and addresses a numerical ambiguity between rotation and translation with a dynamic weighting sub‑problem and alternative search strategy. Experiments on 3D‑2D vascular centerline registration in endovascular image‑guided interventions show processing rates of 60 Hz (without refinement) and 31 Hz (with refinement) on a single‑core CPU, achieving accuracy comparable to existing methods.
By Jingwei Song, Maani Ghaffari
UniReg is a conditional unified model for medical image registration that adapts deformation field estimation based on anatomical priors, registration type constraints, and instance-specific features. It combines the precision of task‑specific learning with the generalization of traditional optimization, enabling effective alignment across diverse CT and MR scenarios within a single framework. Experiments show UniReg outperforms state‑of‑the‑art learning‑based methods in accuracy while providing strong cross‑scenario generalization and reducing training cost and model redundancy.
By Zi Li, Jianpeng Zhang, Tai Ma, Tony C. W. Mok, Yan-Jie Zhou, Zeli Chen, Xianghua Ye, Le Lu, Cheng Chen, Dakai Jin
The paper introduces a privacy‑preserving approach to open‑vocabulary 3D semantic segmentation that operates solely on depth data, eliminating the use of RGB images to avoid disclosing scene‑specific visual information. It proposes a stricter depth‑only evaluation protocol and presents UTTO, a model‑agnostic uncertainty‑guided test‑time optimization framework that refines predictions from frozen open‑vocabulary 3D backbones using structured predictive uncertainty. Experiments on ScanNet and Matterport3D show consistent improvements, and additional analyses demonstrate the method’s relevance for privacy‑constrained robotic applications.
By Xuying Huang, Sicong Pan, Maren Bennewitz
The paper introduces Structured Four-Stage Legal Translation (S4L→Prolog), a reasoning-guided framework that converts raw traffic rules into Prolog logic by performing semantic role extraction, scene completion, logical mapping, and rule generation in a single prompt. Compared to baseline approaches (NL→Prolog and LE→Prolog), S4L achieves higher accuracy—formalizing 75 % of twenty real-world traffic rules versus 60 % and 55 % for the baselines. Qualitative analysis shows S4L better captures implicit causal relations, deontic modality, and exception structures.
By May Myo Zin, Wachara Fungwacharakorn, Ken Satoh, Katsumi Nitta
The paper presents a hybrid generative model that combines denoising diffusion and adversarial training to produce 3D geological microstructures from 2D images. It addresses limitations of previous GAN-based methods like SliceGAN, particularly for heterogeneous materials, by replacing the denoising loss with an adversarial loss to achieve stable training. The resulting model generates microstructures with minimal slice artefacts and accurate phase fractions and structural descriptors.
By Ali Aouf, Eric Laloy, Bart Rogiers, Christophe De Vleeschouwer
The paper argues that traditional probabilistic fairness metrics can miss significant disparities in the actual consequences of decisions. By introducing a utility-based framework, the authors show that a process can satisfy ε-fairness yet still be maximally unfair when utilities are considered. They apply this framework to college admissions and credit‑risk assessment, demonstrating that equalizing probabilities alone may mask unequal utility outcomes across groups.
By Tolulope Fadina, Thorsten Schmidt
The paper introduces SALR, a schema‑anchored latent reasoning approach for generating logical forms in knowledge‑base question answering. SALR delays explicit schema commitments by generating continuous thoughts in hidden states and aligns these thoughts with a codebook of KB schema elements, guided by an alignment objective derived from gold logical forms. Experiments on GrailQA and WebQSP demonstrate that SALR consistently outperforms strong baselines, notably improving compositional question performance by 2.86 F1 points over TIARA.
By Guangze Gao, Zixuan Li, Sikui Zhang, Chunfeng Yuan, Wenjuan Li, Bing Li, Xiaolong Jin, Weiming Hu
The study investigates how limited reading capacity and claim wording influence consensus outcomes in language‑model networks. By modeling message capacity as the number of messages an agent reads, the authors show that when agents read fewer than about 6.4 messages on average, a wrong consensus becomes unreachable. However, the wording of a claim—its inherent threshold—can override this effect, leading to incorrect consensus even when most agents start correct.
By Makoto Fukushima
The paper examines how landslides worldwide are reported in German newspapers, analyzing nearly 55,000 articles covering about 4,500 events over 25 years. It compares media coverage with external measures of countries’ landslide susceptibility, revealing patterns such as the overreporting of Southern and Western Europe. The study highlights spatial biases in media attention to international disasters and suggests avenues for further research on media inequalities.
By Brielen Madureira, Andreas Niekler, Marc Keuschnigg, Mariana Madruga de Brito
The paper introduces Redemption Score (RS), a multi‑modal evaluation framework for image captioning that combines three complementary signals: Mutual Information Divergence for global image‑text alignment, DINO‑based perceptual similarity of cycle‑generated images for visual grounding, and LLM text embeddings for contextual similarity to human references. RS fuses these signals to provide a more holistic assessment, achieving a Kendall‑τ of 58.42 on Flickr8k and outperforming most prior methods. The framework demonstrates consistent performance across Conceptual Captions and MS COCO, offering a robust evaluation that captures both visual accuracy and text quality.
By Ashim Dahal, Ankit Ghimire, Saydul Akbar Murad, Nick Rahimi
Adversarial patches applied to Vision‑Language‑Action (VLA) policies not only corrupt actions immediately but also leave lasting state effects that persist after the patch is removed. The study introduces a state‑restoration protocol that evaluates recoverability after patch removal, distinguishing true adversarial impact from occlusion or action‑error magnitude. Experiments on OpenVLA‑OFT with EDPA attacks show that only 36.2% of episodes recover after five chunks, while controls recover at much higher rates; a recovery adapter can improve recovery but its effectiveness drops with delayed intervention.
By Enhao Wu, Fusen Guo, Yuxin Cao, Ziyang Lyu, Lin Li, Wei Song
The paper introduces Recency Forcing, a technique that addresses the long‑horizon degradation in autoregressive video generation caused by KV eviction mismatch. By applying a timestep‑dependent bias—Temporal Response Bias—derived from a positional response measure, the method reduces the influence of distant frames during inference without altering context length or training objectives. An exact reformulation, Biased Attention Reparameterization, enables this bias to be applied as a standard FlashAttention call with zero overhead, achieving state‑of‑the‑art long‑horizon generation quality on VBench datasets.
By Tri Cao, Hung Nguyen, Phong Nguyen, Khoi Nguyen