The paper introduces the pooled selection-rate standard deviation (pooled SD) as a way to quantify how unevenly multiple-choice option letters are chosen across different orderings. It shows that pooled SD decomposes into two label‑free factors: the flip rate, which measures how often a model’s chosen option changes with ordering, and the positional skew, which captures the variability in option‑letter selection rates. Experiments on 41 checkpoints and 12 benchmarks reveal that prior correction mainly reduces positional skew, while aggregating each ordering with its reversal mainly reduces flip rate; applying both mitigations together further narrows the accuracy range across answer letters.
By Hyunjong Ok, Minjae Park, Jaeho Lee
The paper introduces a multimodal dog‑fingerprinting framework that combines visual identification, multiple‑object tracking, and canine vocal fingerprinting to associate bark events with individual dogs in real‑world videos. It synchronizes bark events with tracked dogs and updates persistent profiles using both visual appearance and vocal characteristics, allowing identification even when dogs are occluded, overlapping vocalizations occur, or dogs bark off‑screen. The authors also present a benchmark of 46 real‑world videos with 213 manually annotated bark events, reporting a 61.83% localization accuracy for visible dogs (IoU ≥ 0.30) and a 91.33% F1 score for detecting visible versus off‑screen barking.
By Aryahvishwa Babu, Kingsley Jin, Lillian Wang, Tuan M. Dang, Theron S. Wang, Jeorge Joshi, Annie Cao, Kenny Q. Zhu
The study audits 16 open RLVR math corpora, totaling 1,673,514 training items, against six benchmarks with 1,949 questions to assess the reliability and data‑loss costs of decontamination policies. Eight policies—based on exact, lexical, and semantic matching—were evaluated using 594 dual‑reviewed training‑benchmark pairs, with calibration and held‑out sets ensuring disjoint benchmark questions and training‑example families. Results show that at a cosine threshold of 0.80, 88% of sampled removals were false, and no semantic threshold in the tested grid met a 90% conservative‑precision target; however, the judge‑adjudicated policy achieved 98.1% resolved precision and 98.2% recall on held‑out corpora, with no confirmed false removals among six sampled removals.
By Esther Xin
The paper introduces a dimension‑free PAC‑Bayes generalization bound for deep neural networks by placing the prior over the non‑parametric space of data distributions using a Dirichlet Process. This shift eliminates the dependence on the number of network parameters, yielding a bound that remains non‑vacuous even for over‑parameterized models. Experiments on MNIST with up to 4.2 million parameters show the new bound staying stable at 8.8 %–9.4 %, whereas classical weight‑space bounds become vacuous.
By Sumit Vashishtha
The paper proposes an AI-native, robust network for sixth-generation (6G) communication systems, where multiple base stations serve extended reality (XR) devices. It introduces a distributed deep learning approach that jointly manages transmit power, computing power, and computation allocation while accounting for uncertainties in channel and computing states. The method uses a deep neural network trained centrally or via federated learning, injecting uncertainty samples after the DNN output to minimize worst-case delay under power and computation constraints, achieving near-centralized performance with improved privacy and scalability.
By Robert-Jeron Reifert, Hayssam Dahrouj, Alaa Alameer Ahmad, Haris Gacanin, Aydin Sezgin
The paper introduces ETH, a training system that implements Heterogeneous Pipeline Parallelism (HPP) to achieve fine‑grained data balance across heterogeneous pipelines. ETH uses a hierarchical workload balance framework and block‑wise gradient aggregation to reduce communication overhead, while also providing optimizations that cut configuration search time. Experiments show ETH can double training speed and cut configuration generation overhead by up to 56× compared to existing systems.
By Mianjie Yu, Zhongmin Zhao, Zizhao Mo, Jianxiong Liao, Wenxuan Li, Yulin Qiao, Huanle Xu, Wenquan Yang, Yongqiang Yang, Zeren Li, Ruifeng Tang, Chengzhong Xu
The paper revisits the Successive Rejects (SR) algorithm for best‑arm identification, showing that although SR appears non‑adaptive due to fixed phase lengths, it achieves an optimal worst‑instance error exponent for Gaussian arms with common variance. The authors formulate an optimization problem to design a phase schedule that maximizes the normalized error exponent, leading to a new variant called H1‑balanced Successive Rejects (H1‑SR). H1‑SR attains at least 11/12 of the optimal normalized rate for all K ≥ 2, approaching the optimum as K grows, and experiments demonstrate significant performance gains over the original SR while highlighting trade‑offs in early elimination strategies.
By Mohamed Khair, Junpei Komiyama
SurvICL is a tabular foundation model designed for in-context survival analysis, predicting survival curves directly from data. It is pretrained on synthetic datasets generated via a novel flexible proportional‑odds prior and trained using a plain quantile regression loss, enabling it to handle uncensored event times. On a new benchmark of 103 survival datasets, SurvICL outperforms 20 baseline models—including the classic Cox model—without any tuning, achieving superior integrated Brier scores on 63 held‑out datasets.
By Julie Alberge, Linus Bleistein, Judith Ab\'ecassis, Marine Le Morvan, Ga\"el Varoquaux, David Holzm\"uller
The paper presents a hybrid forecasting model for tourist arrivals to Hong Kong and Sanya that blends Internet big data, a cooperative game‑theoretic weighting of social‑media opinion leaders, Empirical Wavelet Transform for non‑stationarity, and a Bayesian‑optimized Stacked Bi‑LSTM network. It demonstrates that incorporating trust‑weighted social media signals reduces one‑step‑ahead root mean squared error by 4.1–41.7 % for Hong Kong markets and 5.3 % for Sanya, with additional opinion‑leader weighting improving accuracy by 10.1–49.4 % and 8.7 % respectively. The authors prove closed‑form Shapley values and polynomial‑time coalition identification, and show the method outperforms published benchmarks under volatile conditions.
By Ali Nikseresht
The paper introduces Thinking-Free Policy Initialization (TFPI), a simple adaptation to Reinforcement Learning with Verifiable Reward (RLVR) that removes explicit thinking content during training and inference. TFPI bridges long Chain-of-Thought distillation with standard RLVR, reducing token usage and improving performance. Experiments show that TFPI accelerates RL convergence, raises performance ceilings, and yields token‑efficient reasoning models, enabling a 4B model to achieve 89.0% accuracy on AIME24 and 65.5% on LiveCodeBench with under 4K H20 hours.
By Xin Xu, Clive Bai, Kai Yang, Tianhao Chen, Yang Wang, Saiyong Yang, Can Yang
Neuroscience data are fragmented across labs and formats, making reuse labor‑intensive. This study benchmarks coding agents on eight recent mouse neural recording papers, asking them to reformat data for a common decoder task. While agents handle many subtasks, they rarely produce fully error‑free solutions; the authors pair outcome metrics with a process rubric to evaluate errors and propose agent‑assisted data‑sharing practices.
By Ling-Qi Zhang, Kristin Branson
The paper introduces a multimodal emotion recognition system that integrates facial expressions, speech, text, and body movement to aid clinicians in psychological evaluation. While individual modalities perform well on benchmarks, their real‑world accuracy drops to 82.97%–87.64%, whereas the fused multimodal system reaches 96.43% accuracy with no state below 92.31%. These findings demonstrate that multimodal fusion enhances robustness in realistic settings, supporting more reliable evaluator assistance.
By Kris Kraack
A new distribution-to-distribution (D2D) neural probabilistic forecasting framework is introduced, using kernel mean embeddings to encode input distributions and mixture density networks to parameterise output distributions. The framework allows recursive multi-step forecasting directly in distributional form, avoiding the need for explicit ensemble simulations. Demonstrated on the Lorenz63 system, the D2D model captures complex distributional evolution and achieves skillful probabilistic forecasts comparable to a perfect-model benchmark.
By Tianlin Yang, Hailiang Du, Louis Aslett
The paper introduces Mentalized Adversarial Self-Play (MASP), a framework for training proactive dialogue agents by modeling the user as a rationally resisting participant with an evolving mental state. MASP separates teacher and student mentalization, alternates self‑play updates for system and user policies, and treats robustness to rational resistance as a training objective. Experiments on proactive dialogue benchmarks show MASP improves success rate, rationality‑constrained success, reduces average turns, and performs better against hard users compared to SFT and Planner‑only approaches.
By Zanwu Liu, Dan Zhang, Xiaodi Wang, Yongbo Mao, Xiaoyang Guo, Yonglin Wu, Peiyue Jiang, Yangyang Wei, Zecheng Lin
The paper surveys memory‑centric compression techniques for large language models, framing long‑context handling as a memory‑management problem. It distinguishes compute‑active memory from compute‑inactive memory, and models operations such as writing, offloading, recall, and forgetting as state transitions within an information lifecycle. The authors propose a unified cost model and argue that future research should focus on dynamic, lifecycle‑aware memory management rather than static compression.
By Chaogui Gou
Hy‑Memory‑30B‑A3B is a compact long‑term‑memory writer that learns to write to external memory across multiple frameworks (Mem0, MemOS, and Hy‑Memory). It uses Memory On‑Policy Distillation to teach framework‑specific write protocols and Memory RL to optimize answerability with dense per‑chunk probes, attributed cross‑chunk rewards, and chunk‑level GRPO. On a 20% benchmark subsample, it outperforms other models, achieving a 55.49% macro average versus 54.62% for Claude Opus 5.
By Tianxiang Fei, Jiazheng Kang, Mao Zheng, Xingyan He, Xiang Yu
Grid‑Bench is an executable benchmark that tests large language models’ ability to predict the final state of discrete worlds after a sequence of spatial operations. The benchmark uses two‑, three‑, and four‑dimensional grids and provides explicit rules, an initial grid, and a textual event history for the model to output the complete final grid. Evaluation of six LLMs shows that accuracy drops sharply with longer histories and that models with similar exact‑match scores can differ widely in the number of incorrect cells, highlighting nuanced performance differences.
By Zixiang Xu, Zirui Song, Wanrong Zheng, Chiyu Zhang, Yue Huang, Yuyang Peng, Kexin Zhang, Jinkun Chen, Xiuying Chen, Xiangliang Zhang, Yan Liu, Laurent Itti, Frank Shyu, Xiaoqiang Luo, Sandeep Pandey, Xi Liu, Luke Simon, Xiyang Hu, Yue Zhao
VELD-QA is a large-scale benchmark for visual evidence grounding and question answering in long documents, comprising 15,000 human-authored question-answer pairs across 5,859 PDFs with 372,445 pages. The dataset includes detailed annotations of supporting pages, visual evidence elements, and descriptions of cited visual content, enabling separate evaluation of retrieval, localization, and interpretation. Experiments on nine vision-language models and various retrieval-augmented systems reveal significant gaps in evidence localization and visual interpretation, with oracle accuracy peaking at 81.7% and enrichment of visual-element annotations narrowing performance gaps by 17–24%.
By Zhivar Sourati, Marianne Menglin Liu, Rongguang Wang, Zheng Wang, Yazhe Hu, Lan Guo, Yuying Wang, Mengqing Guo, Bhavesh Dnyaneshwar Gore, Shubham Pathak, Ramapriya Ranganath, Sandeep Sharma, Tao Sheng, Sujith Ravi, Dan Roth
EVA-Cap is a new framework for audiovisual video captioning that breaks captions into atomic audiovisual events, turning unstructured text supervision into event‑centric alignment. The model is trained with supervised fine‑tuning and Group Relative Policy Optimization on a self‑curated dataset, EVA‑Data, which applies event‑centric quality control. Experiments show EVA‑Cap outperforms prior methods on event‑centric metrics (e.g., +9.0% on ChronusAV) while remaining competitive on general captioning benchmarks (e.g., +3.3% on WorldSense).
By Jiapeng Shi, Qiuxia Hou, Jiaqi Leng, Junke Wang, Yanhao Zhang, Haonan Lu, Ziyi Ye, Zuxuan Wu
The paper introduces CAPRI, a dataset of conversations designed to test whether large language models (LLMs) can adapt their responses to users’ cultural backgrounds. Experiments show that while LLMs can infer a user’s cultural context and recall relevant conventions, they frequently fail to apply this knowledge when generating answers unless explicitly prompted to do so. The study also finds that LLMs’ cultural adaptations improve as more cues are provided, yet their default priors are not culture‑neutral and sometimes reflect the model’s country of origin.
By Yisong Miao, Jian Zhu, Vered Shwartz