PolERo presents a new dataset of 3,574 Romanian question‑answer pairs from presidential transcripts, annotated for political evasion using a two‑level taxonomy of response clarity and fine‑grained evasion strategies. The study evaluates various classification methods—including TF‑IDF baselines, fine‑tuned encoders, a sliding‑window encoder, and zero/few‑shot LLM prompting—under matched conditions. Cross‑lingual transfer experiments via joint bilingual training and machine‑translation augmentation reveal that fine‑tuned encoders perform competitively, transfer is asymmetric, and ambivalent evasion categories with pragmatic cues remain the most challenging across all models.
By Gabriel Stefan, Sergiu Nisioi
The paper introduces CoSMo-RecNet, a modular framework for multi-contrast MRI reconstruction that operates effectively in low-data regimes. It leverages a reusable content/style prior learned from large, unpaired multi-contrast image datasets, allowing a lightweight unrolled network to refine reconstructions using only a few task‑specific training samples. Experiments on low‑field 0.3 T and ultra‑low‑field 47 mT datasets demonstrate that CoSMo-RecNet outperforms parameter‑matched MoDL, classical reconstruction, transfer learning, and zero‑shot methods, achieving higher quality with as few as five training subjects.
By Chinmay Rao, Efe Il{\i}cak, Matthias J. P. van Osch, Mariya Doneva, Laurens Beljaards, Navid Jabarimani, Nicola Pezzotti, Marius Staring
The paper proposes a method to adapt the Depth Anything V2 (DAV2) zero‑shot relative depth model for estimating lunar surface height. By fine‑tuning DAV2 with publicly available stereophotogrammetry‑derived DEM data, the authors achieve a significant performance boost over the unadapted zero‑shot model. This improved estimator can provide more accurate relative height information useful for hazard detection in future ESA lunar landings.
By Patrick Bauer, Marius Schwinning, Melanie Siegel, Andreas Weinmann, Hichem Snoussi
The paper introduces Inference‑Native Zeroth‑Order (ZO) optimization, which redefines ZO as a query‑based process that can be executed directly by inference runtimes. By exposing ZO’s query semantics and using abstractions such as ProbePlan, factorized side states, and persistent subspace reuse, the method reduces state‑management cost and DRAM traffic dramatically. Experiments on large models (OPT‑13B, Qwen3‑8B) show that inference‑native steps are nearly identical to matched‑query controls while achieving significant memory savings and efficient batching.
By Zelin Li, Caiwen Ding
The paper introduces a two‑stage framework for recognizing depression symptoms at the sentence level. First, a contrastively fine‑tuned sentence encoder generates a symptom candidate for each sentence. Then, a fine‑tuned language model verifies the candidate’s presence or absence by comparing the sentence, its context, and a diagnostic definition, ensuring the model’s judgment aligns with that definition before responding.
By Weiming Li, Catarina Barata, Miguel Constante, Joao Sanches
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
PlantC2USeg is a deep transfer‑learning framework that uses cross‑scale consistency learning and an information‑restricted decoder to improve plant point cloud segmentation. It achieves state‑of‑the‑art performance on Soybean3D and ShapeNet Part, and demonstrates strong few‑shot generalization across species and sensing conditions. The method reduces the need for large annotated datasets and lowers adaptation overhead for new plant species.
By Yu Tian, Xintong Jiang, Jan Franklin Adamowski, Shiv O. Prasher, Shangpeng Sun
The paper investigates how Large Language Models (LLMs) construct fictional worlds, specifically examining setting as a measurable aspect of storyworld creation. By generating 1,000 AI stories per model in English and German and comparing them to human-authored fiction from Project Gutenberg, the authors classify narrative space into five categories—action, perceived, visual, descriptive, and no space—using fine‑tuned BERT classifiers. Results show that human texts mainly use action space, grounding narratives in character-environment interaction, while LLMs consistently overproduce perceived space, focusing on atmosphere and affect, with this pattern varying by model and language.
By Katrin Rohrbacher, Bj\"orn Nieth, Emmanuelle Salin, Bjoern Eskofier, Michaela Mahlberg
GDB-Reward is a framework that converts diverse graphic design evaluation metrics into a single reinforcement learning reward, enabling prompt optimization for frozen image generators. By using these metrics as scalar rewards, the method improves adherence to design specifications in perceptual quality, rendering fidelity, and spatial accuracy without fine‑tuning the generator. Experiments show that non‑differentiable metrics can serve as effective optimization objectives in settings lacking differentiable supervision.
By Adrienne Deganutti, Purvanshi Mehta, Simon Hadfield, Andrew Gilbert
The paper introduces a new approach to low‑rank clone distillation that ensures the student model’s training targets exactly the weights it will use at inference. By redefining the training objective to cover the full deployed matrix—without changing the model’s shape, parameter count, or FLOPs—the authors recover previously unreachable linear degrees of freedom. This results in significant performance gains across multiple teacher models, achieving comparable or superior accuracy with fewer tokens and parameters.
By Wenhui Chen, Zhifeng Li, Jie Zhou, Navan Preet Singh, Madalina Ciobanu, Chenghua Wang, Qingqing Mao, Ritankar Das
SocialMaze is a new benchmark designed to evaluate large language models on social reasoning tasks that involve deep reasoning, dynamic interaction, and information uncertainty. It comprises six tasks drawn from social deduction games, everyday interactions, and digital communities, and includes automated checks and human validation to ensure data quality. Experiments with twelve LLMs reveal that stronger chain‑of‑thought reasoning improves performance on deeper inference tasks, while uncertainty consistently hurts results; targeted fine‑tuning on curated reasoning traces can markedly enhance structured social‑reasoning abilities.
By Zixiang Xu, Yanbo Wang, Yue Huang, Haomin Zhuang, Yujun Zhou, Jiayi Ye, Sixian Li, Zirui Song, Lang Gao, Chenxi Wang, Zhaorun Chen, Wang Pan, Yue Zhao, Jieyu Zhao, Xiangliang Zhang, Xiuying Chen
The study evaluates fundus-specific foundation models (FM) for detecting diabetic macular edema (DME) in retinal images. It compares two popular FM—RETFound and FLAIR—against a lightweight EfficientNet-B0 backbone across multiple datasets (IDRiD, MESSIDOR-2, and OCT-and-Eye-FundusImages). Results indicate that FM do not consistently outperform fine‑tuned CNNs; EfficientNet-B0 often matches or exceeds FM performance, with FLAIR being the most competitive FM.
By Franco Javier Arellano, Jos\'e Ignacio Orlando
Task-Level Natural Language Priors as Learning Signals for Low-Resource LLM Training proposes Prior-Guided Tuning (PGT), a training approach that treats natural-language priors as auxiliary learning signals rather than just input context. The method introduces Contrastive Prior Steering (CPS), which adds positive and negative prior-conditioned auxiliary losses while preserving the original supervised objective. Experiments on AmbiMath, Jigsaw, and MNLI/HANS demonstrate that CPS consistently outperforms plain and prompt fine-tuning, achieving high accuracy and significant gains with limited training data.
By Jian Gao, Xiao Zhang, Xun Zhu, Miao Li, Ji Wu
This study experimentally evaluates large language models (LLMs) for legal decision-making in Austrian and EU value‑added tax (VAT) law. It compares fine‑tuning and retrieval‑augmented generation (RAG) on textbook and real‑world tax consulting cases to identify optimal LLM configurations and assess legal‑reasoning abilities. Results show that properly configured LLMs can aid tax professionals with routine tasks and provide legally grounded justifications, though they are not yet ready for full automation due to challenges like hallucination and handling implicit client knowledge.
By Marina Luketina, Andrea Benkel, Christoph G. Schuetz
UniToolCall introduces a unified framework for tool-use in large language model agents, standardizing toolset construction, dataset generation, and evaluation. The framework aggregates over 22,000 tools and creates a hybrid training corpus of more than 390,000 instances by combining ten public datasets with synthetically generated, structurally controlled trajectories. It models diverse interaction patterns—single‑hop vs. multi‑hop, single‑turn vs. multi‑turn, serial vs. parallel execution—and adds an Anchor Linkage mechanism to enforce cross‑turn dependencies, while converting seven public benchmarks into a common Query–Action–Observation–Answer format for fine‑grained evaluation.
By Yijuan Liang, Xinghao Chen, Yifan Ge, Ziyi Wu, Hao Wu, Changyu Zeng, Wei Xing, Xiaoyu Shen
XMerge is a post‑training method for compressing large language models by removing entire transformer layers while preserving a standard serving architecture. It selects low‑impact blocks via cross‑axis selection and refits adjacent surviving blocks with local boundary reconstruction, requiring no task labels or fine‑tuning. Across seven Llama and Qwen backbones, XMerge outperforms five published baselines, achieving top rankings on CORE and MMLU tasks even at aggressive compression levels, and consistently avoids model collapse while improving calibration and decoding efficiency.
By Jundong Hu, Shekar Ramachandran
VideoPulse is a neonatal dataset and end‑to‑end pipeline that estimates heart rate and peripheral capillary oxygen saturation (SpO2) from facial video without contact. The dataset contains 157 recordings from 52 neonates, and the pipeline uses face alignment, artifact‑aware supervision, and 3D CNN backbones to produce predictions every 2 seconds. On the NBHR dataset the model achieves a heart‑rate MAE of 2.97 bpm and SpO2 MAE of 1.69 %.
"whyItMatters":"The results show that short, unaligned neonatal video segments can accurately estimate vital signs, offering a low‑cost, non‑invasive monitoring option for neonatal intensive care."
By Deependra Dewagiri, Kamesh Anuradha, Pabadhi Liyanage, Helitha Kulatunga, Pamuditha Somarathne, Udaya S. K. P. Miriya Thanthrige, Nishani Lucas, Anusha Withana, Joshua P. Kulasingham
The paper introduces a zero‑shot subject‑driven video generation framework that eliminates the need for per‑subject tuning and large subject‑video datasets. It achieves this by separating identity injection—learned from subject‑image pairs—and motion‑awareness preservation—maintained with a small set of arbitrary videos, and optimizes both with stochastic switching and dropout techniques. Using CogVideoX‑5B, the method adapts a single model with only 200K subject‑image pairs and 4,000 arbitrary videos in 288 A100 GPU hours, representing roughly 1% of the compute required by previous zero‑shot baselines while preserving subject fidelity and motion quality.
By Daneul Kim, Jingxu Zhang, Wonjoon Jin, Sunghyun Cho, Qi Dai, Jaesik Park, Chong Luo
The paper presents a fully automated pipeline that extracts maize ear traits—such as kernel count, row number, and kernel size—from 3D point clouds generated by a video-to-point-cloud platform. The method processes raw video through COLMAP and NeRF, isolates the ear, calibrates the point cloud, aligns it, unwraps it into a 2D image, and applies Cellpose‑SAM for instance segmentation, achieving high accuracy (kernel count R² = 0.921, MAPE = 10.33 %) on a held‑out dataset. The resulting multi‑trait dataset, with genotype identities, is ready for phenotype‑to‑genotype association studies.
By Ritwesh A. Kumar, Som Tripathi, Peja Matthews, Srikar Reddy, Talukder Zaki Jubery, Patrick Schnable, Adarsh Krishnamurthy, Baskar Ganapathysubramanian
TrajMind is a framework for diagnosing collective anomalies in urban trajectory data. It separates continuous screening from on-demand diagnosis, using a fast text-only path for alerts and a slow vision‑language path that chains role‑specialized LoRA adapters for detailed, evidence‑backed what‑who‑where‑when records. Experiments show the slow path outperforms baselines by over 15 percentage points in typing and 13 in localization, while the fast path cuts latency by 41% and retains high accuracy.
By Jiahao Wu, Zhenqun Yang, Chen Jason Zhang, Qing Li