The study applies contrastive representation learning (CEBRA) to EEG data collected from conversational dyads, embedding the recordings onto a 2‑dimensional sphere. The resulting manifold captures individual identity features, as evidenced by high decoding accuracy for dyad labels and significant differences when labels are permuted only after retraining the encoder. However, the embedding shows little sensitivity to conversational role (speaker vs. listener) and to autism‑quotient differences, suggesting that the geometry is dominated by individual traits rather than task‑specific dynamics.
By Hubert Huang, Michelle McCleod, Brendan Ames, Evie Malaia
HakemBench is a Turkish benchmark for typed decision-making tasks, comprising 2,346 items and 4,275 choice, yes/no, and score questions across seven tracks (fact‑check triage, education, guardrails, legal routing, moderation, spam and phishing, and customer support). The benchmark evaluates models on decision quality (macro F1), calibration (normalised Brier score), and selective automation (normalised area under the generalised risk‑coverage curve), combining these metrics by a geometric mean and reporting confidence intervals from 2,000 bootstrap draws. Gold labels are generated from blind passes of a single AI model family compared with votes from other large language model families, and the benchmark’s leader achieved a composite score of 0.888 while the lab’s own model ranked 7th with 0.660.
By Sait Furkan Teke (ufak AI)
FiberGeoText (FGT) is a vision‑language model that clusters short‑range superficial white matter streamlines from ultra‑high‑resolution diffusion MRI into population‑level groups. It jointly encodes each streamline’s 3‑D trajectory, cortical anatomical context (via text from multiple parcellation schemes), and shape, using a pretrained large language model to unify heterogeneous anatomical descriptions. Evaluations on 0.76 mm diffusion data show that FGT outperforms state‑of‑the‑art methods in cortical parcel coherence, shape consistency, cluster‑size consistency, and cross‑subject correspondence, and it generalizes well to unseen subjects, recovering 96.7 % of learned clusters.
By Yuqian Chen, R. Jarrett Rushmore, Guikun Chen, Fan Zhang, Edward Yeterian, Nikos Makris, Yogesh Rathi, Lauren J. O'Donnell
The paper presents a new retrieval benchmark for the model organism Dictyostelium, created from curator-generated queries linked to PubMed articles and gene annotations. It evaluates three niche retrieval strategies—cross‑encoder reranking, gene‑aware query expansion, and abstract‑only versus full‑text retrieval—finding that reranking and gene‑aware expansion improve results in specific contexts, while full‑text chunks boost recall and top‑rank performance when abstracts lack supporting evidence. The benchmark, data, and code are publicly released on GitHub and Zenodo.
By Yun Wang, Gad Shaulsky, Toma\v{z} Curk, Bla\v{z} Zupan
The paper introduces OrbFlow, an SE(3)-equivariant generative model that predicts Gaussian-type orbital coefficients for electron density using flow matching. By training with a two-phase trajectory curriculum, OrbFlow improves accuracy on QM9 and MD benchmarks, reducing density errors significantly compared to previous models. The predicted densities also accelerate self-consistent field procedures, cutting iterations by up to 68% and accurately reproducing dipole and quadrupole moments without additional SCF calculations.
By Chenxing Liang, Chengdong Wang, Yuchao Lin, Xiaofeng Qian, Shuiwang Ji
The paper introduces Test-Time Calibration Learning (TTCL), a label‑free framework that adapts both reasoning accuracy and verbalized confidence of large language models directly on unlabeled target‑task data. TTCL generates self‑supervision signals from multiple model responses, enabling calibration without ground‑truth labels and proving theoretically as a bounded surrogate for the ideal calibration objective. Experiments on mathematical reasoning and factual question answering show consistent improvements, with base models gaining an average 40.13% in accuracy and 70.80% in ECE reduction across eight benchmarks, and further gains under domain shift.
By Zizhuo Zhang, Xiong Peng, Jingwei Sun, Rong Yao, Shixiong Kai, Mingxuan Yuan, Bo Han
The paper introduces FOVEATED, a plug‑and‑play framework that improves atomic‑fact recall in unstructured knowledge editing (UKE) for large language models. By randomly shifting Rotary Position Embedding (RoPE) positions during editing, FOVEATED creates focused views of each sentence, counteracting the context‑reliance problem where edited LLMs reproduce passages but fail to recall individual facts. Experiments show consistent gains across five editors, two LLM backbones, and three benchmarks.
By Ding Wu, Ye Zhang, Haoyu Wang, Tianci Liu
The paper introduces Causal-Invariant Masking (CIM) to better quantify epistemic uncertainty in Multimodal Large Language Models (MLLMs) by measuring semantic shift between original predictions and those conditioned on a causally-focused view. It proposes Semantic Divergence as a core metric that converges to the variance of the model’s sensitivity to non‑causal correlations, and introduces Expected Embedding Drift (EED) as a fast geometric proxy that estimates this shift directly in the embedding space. Experiments demonstrate state‑of‑the‑art uncertainty quantification performance and a nearly 50% speedup with EED.
By Haoyang Luo, Linwei Tao, Jie Gui, Xinghao Chen, Chang Xu, Jianyuan Guo, Minjing Dong
The paper introduces the concept of world editing—intervening on an existing executable world while preserving unchanged properties—and defines intervention depth as a measure of how strongly edits affect world entities, dynamics, and systems. It presents IGMWorld and IGMBench, a benchmark of 110 tasks across Minecraft and Terraria, evaluating edits through deterministic executability, behavioral, preservation, and visual checks. Frontier coding agents achieve up to 78.2% task-level success and 94.8% criterion-level success, though reliability drops with deeper interventions and visual consistency remains low.
By Max Ku, Nok-Kan Law, Yu-Chien Tang, Shih-Ying Yeh, Ping Nie, Andy Zheng, Tat Hei Lai, Fei-Yueh Chen, Nikko Yu, Wei-Chieh Sun, Suzy Huang, Chiao-Wei Hsu, Chih-Chuan Huang, Chak-Wing Mak, Ho Yin Sam Ng, Edisy Kin Wai Chan, Min-Hung Chen, Ho Kei Cheng
Enrich-on-Graph (EoG) is a flexible framework that uses large language models to enrich knowledge graphs, thereby bridging the semantic gap between structured graphs and unstructured queries in complex reasoning tasks. By leveraging LLMs’ prior knowledge, EoG enables efficient evidence extraction from knowledge graphs, achieving precise and robust reasoning while maintaining low computational costs and scalability. The authors also introduce three graph quality evaluation metrics for query‑graph alignment, theoretically validate their optimization objectives, and demonstrate state‑of‑the‑art performance on two KGQA benchmark datasets.
By Songze Li, Zhiqiang Liu, Zhengke Gui, Huajun Chen, Wen Zhang
LiteEMG-FM is an efficient hybrid CNN‑Transformer foundation model designed for electromyography (EMG) sensing. It is pretrained on 16 diverse upper‑ and lower‑limb EMG datasets, enabling representations that generalize across users and datasets. The model incorporates a lightweight, always‑on 1D‑CNN wake‑up module that filters rest and non‑target activity, activating LiteEMG‑FM only for valid gestures, and is evaluated in full inference offloading, split inference, and full on‑device processing scenarios, showing superior performance over state‑of‑the‑art time‑series foundation models and supervised baselines, especially in zero‑calibration cross‑participant and data‑scarce conditions.
By Tianhao Wu, Xu Wu, Amirmohammad Radmehr, Jiawei Yu, Yi Wu, Phuc Nguyen, Jian Liu
SoftGene is a new framework that enhances gene set annotation by combining protein language model representations with a hybrid prompting scheme. It uses a hierarchical attention-based encoder built on ESM to encode gene sets from protein amino acid sequences, then merges soft prompts derived from these embeddings with hard prompts generated by a large language model. The approach is evaluated on Gene Ontology and MSigDB datasets, showing that integrating protein-sequence information with textual context improves overall annotation performance, though the benefit varies across biological domains.
By Drew Ross, Arya Hadizadeh Moghaddam, Dongjie Wang, Xiaoyu Zhang, Zijun Yao
The article introduces two new authorship attribution metrics, Rank‑Turbulence Delta and Jensen‑Shannon Delta, which extend Burrows’s classical Delta by using distance functions suited to probabilistic distributions. It explains the theoretical foundations, contrasts centred versus uncentred z‑scoring, and presents a token‑level decomposition that makes each Delta distance numerically interpretable. The methods are evaluated on four multilingual literary corpora, showing that Rank‑Turbulence Delta matches Cosine Delta in accuracy while Jensen‑Shannon Delta often outperforms the traditional Delta, and the study also reassesses existing attribution algorithms on a large Russian benchmark.
By Dmitry Pronin, Evgeny Kazartsev
The paper introduces a Fisher-guided submodular data selection method for continual pre‑training of large language models, addressing the forgetting problem that arises when a target‑domain corpus overwrites pretrained knowledge. By decomposing candidate gradients into anchor and frontier components and optimizing a log‑determinant submodular objective, the method selects data that both improves target‑domain performance and limits forgetting. Experiments on TinyLlama‑1.1B and Llama‑3.1‑8B show that the selector achieves superior adaptation and forgetting control while using ten times fewer tokens than traditional replay strategies.
By Zhenghao Zhao, Gaowen Liu, Zhiling Lan, Yan Yan
The paper introduces SLIM, a pushing benchmark that tests language‑based goal specification for latent world models. It shows that a vanilla LeWM fails almost entirely due to an action‑insensitive encoder, but adding an inverse‑dynamics auxiliary loss restores encoder sensitivity and dramatically improves success rates. With the repaired latent, a lightweight language‑goal head can plan from natural language sentences, achieving high performance on navigation and pushing tasks without retraining the world model.
By Florian Strohm, Patrick Wagner, Jannik Schwab, Marco Huber
The paper compares two low‑budget methods for converting a 30B Mixture‑of‑Experts autoregressive language model into a diffusion language model. One method updates a subset of the model’s weights in‑place, while the other freezes the context tower and conditions on a frozen causal copy via cross‑attention. With only 1B training tokens, the frozen‑tower approach achieves a HumanEval pass@10 score of 71.60 versus 6.19 for the in‑place method, and retains 95% of the parent’s GSM8K and 99% of its MMLU‑Pro performance.
By Wentao Lu, Jesse Clark, Tianyu Zhu
LPS-Bench is a benchmark designed to evaluate the safety awareness of computer‑use agents (CUAs) in long‑horizon planning tasks that involve tool workflows. It uses a template‑guided multi‑agent pipeline to generate user instructions, simulated toolkits, and case‑specific safety criteria, followed by human review, allowing scalable expansion without building separate application environments. The benchmark includes 570 cases from 65 scenarios across seven task domains and nine planning‑risk types, and an LLM‑based evaluator assesses tool choices, arguments, and responses throughout execution. Evaluations of 13 LLM agents show persistent safety failures in both benign and adversarial settings, with prompt‑based interventions providing only model‑dependent improvements.
By Tianyu Chen, Chujia Hu, Dongrui Liu, Xia Hu, Wenjie Wang
SyntaxBench is a diagnostic benchmark and statistical evaluation framework for character‑level reasoning in large language models, comprising five core tasks—character counting, letter containment, palindrome detection, edit distance, and longest‑string selection—and a harder substring‑extraction stress test called index_to_span. The benchmark uses paired English and random‑string inputs, zero‑, one‑, and four‑shot prompts, and evaluates models from 2B to 32B parameters across multiple reasoning modes. It reports a wide range of metrics, including exact‑match and relaxed accuracy, Cohen’s kappa, McNemar tests, bootstrap confidence intervals, Kendall’s tau, class‑conditional metrics, tokenization analysis, and multiple‑comparison‑corrected tests.
By Mohsen Larni (Department of Computer Science, University of Nevada, Las Vegas), Sobhan Ebrahimi Azar (Department of Computer Science, University of Nevada, Las Vegas), Pouyan Nahed (Department of Computer Science, University of Nevada, Las Vegas), Kazem Taghva (Department of Computer Science, University of Nevada, Las Vegas)
The paper introduces FaVOS, a benchmark for Video Object Segmentation (VOS) that focuses on scenarios where target objects appear only intermittently over long videos. It demonstrates that the standard J&F metric can be gamed by empty predictions, allowing trivial models to outperform strong ones like SAM 3. To address this, the authors propose Volumetric J&F, which treats mask sequences as spatio‑temporal volumes, reducing the influence of target‑absence rewards while maintaining sensitivity to segmentation quality and temporal structure.
By Jihwan Hong, Woohyeon Park, Jaeik Kim, Jaeyoung Do
The paper introduces PSPO, a method for offline policy optimization that treats the dynamics model as a random variable, enabling controlled exploration of out‑of‑distribution (OOD) regions while maintaining robustness. By alternating updates of the posterior distribution and the policy, PSPO achieves a regularized optimization algorithm with convergence guarantees. Experiments on standard benchmarks show that PSPO outperforms state‑of‑the‑art baselines and retains a pessimism‑free property without sacrificing robustness.
By Hongqiang Lin, Dongxu Zhang, Yiding Sun, Mingzhe Li, Ning Yang, Haijun Zhang