The paper introduces a new framework that uses 2D Gaussian Splatting to capture view‑dependent specular effects, especially near‑field reflections, by directly leveraging observable environment information in input images. It combines a hybrid reflection appearance representation with view‑dependent specular modeling, and demonstrates improved rendering quality and geometric accuracy on public benchmarks and a newly curated near‑field reflection dataset.
By Yiming Liang, Tianyu Xiao, Hiroshi Ishikawa
The paper presents a new multi-frame representation for event-based action recognition that outperforms single-frame approaches, especially on long and complex event sequences. It employs a masked auto-encoding framework to pre-train a Vision Transformer, using a sparsity-aware masking algorithm, token selection, and multiple mask tokens with position-aware attention to boost performance while managing computational cost. The method achieves state‑of‑the‑art results on the THU‑EACT‑50 CHL and DailyDVS‑200 benchmarks.
By Emma Norrby, Tomas Wilkinson, Per Nyblom, David Gustafsson
The paper introduces CASTER, a framework that formalizes the conversion of synthetic videos into real robot trajectories as a constrained optimization problem. It shows that synthetic videos must probabilistically seed the search within a semantic basin of attraction, allowing local trajectory refinement while maintaining task alignment. By removing simplifying assumptions from existing methods, the authors achieve more robust zero‑shot manipulation and improve task success by 34.1% over the best baseline.
By Yinlong Dai, Homanga Bharadhwaj, Dylan P. Losey
The paper introduces Motion Estimation-based Video Convolution (MEVC), a framework that removes the need for image signal processors by operating directly on raw Bayer pixel data. MEVC incorporates sliding‑window motion estimation into each convolutional layer, allowing prediction and residual refinement to cut redundant computations across frames. The resulting end‑to‑end pipeline achieves more than 70% reduction in FLOPs while maintaining accuracy on video semantic segmentation, depth estimation, and object detection tasks.
By Haichao Wang, Jiangtao Wen, Yuxing Han
SAGE-OR is a semi‑supervised framework for generating scene graphs in operating rooms that eliminates the need for dense multi‑modal supervision and specialized hardware. It replaces the traditional detect‑then‑reason pipeline with a decoupled representation‑reasoning approach, using frozen foundation models for localization and a lightweight 15‑million‑parameter graph transformer for relational reasoning. On the 4D‑OR benchmark, the core model matches fully supervised baselines (76% F1) and, with unsupervised hand augmentation, reaches 86% F1, achieving near state‑of‑the‑art performance while requiring only relationship and class labels.
By Brandon Leblanc, Charalambos Poullis
The paper reports on the updated AlgoPerf V1 benchmark, evaluating new training algorithms on wall‑clock time‑to‑target across nine workloads on fixed hardware. Using a self‑tuning ruleset that requires a single configuration for all workloads, Schedule‑Free AdamW achieved the best overall performance among 13 PyTorch and JAX submissions. The study also examines trade‑offs between step efficiency and runtime, finding that matrix‑preconditioned and two‑loop methods are sample‑efficient but need aggressive computational optimizations, and highlights the high sensitivity of learning rate and momentum across domains, underscoring the need for more robust hyperparameter‑tuning‑resistant methods.
By Priya Kasimbeg (Alice), Ahmed Khaled (Alice), Aahlad Chandrabhatta (Alice), Yifan Wang (Alice), Andy Liu (Alice), Niccol\`o Ajroldi (Alice), David Tweedle (Alice), Xingyu (Alice), Yang, Isaac Patole, Frank Schneider, George E. Dahl
The paper investigates how different calendar feature encodings influence the performance of covariate-informed time series foundation models (TSFMs). Seven encoding schemes—three target‑agnostic (ordinal, one‑hot, cyclic) and four target‑based—are systematically compared across four state‑of‑the‑art TSFMs on nine datasets that reflect social activity patterns such as weekends and holidays. Results show that encoding choice can change forecast accuracy by up to 9.9 % on non‑holidays and 83 % on holidays, with improvements mainly around holidays and varying effects across models.
By Alexander Kreusel, Maximilian Beichter, Alexandra Nikoltchovska, Benjamin Sch\"afer, Ralf Mikut, Frederik Zahn, Veit Hagenmeyer
QTestBench is a new framework that evaluates quantum code tests generated by large language models (LLMs) by comparing their verdicts against independent benchmark checkers, rather than treating the tests as ground truth. It prompts models to produce syntactic, semantic, and behavioral tests, injects candidate solutions at runtime, and assesses test quality across three validation cohorts. In benchmarks on PennyLane QHack challenges and Qiskit HumanEval tasks, generated test suites frequently execute but often fail to discriminate correct from faulty implementations, with the best test quality scores of 0.71 and 0.34 respectively.
By Meriem Terki, Nouhaila Innan, Minghao Shao, Muhammad Kashif, Alberto Marchisio, Muhammad Shafique
CLEAT (CLarification through valuE-guided Agent co-Training) is a framework that jointly trains a clarification-value controller and a task policy to improve interactive intent alignment in large language model agents. It introduces a residual query value metric, estimated via paired rollouts, to assess the utility of asking clarification questions and uses this to guide both controller predictions and policy decisions. Experiments with Qwen3-4B show CLEAT outperforming baselines across multiple benchmarks, with significant gains in user‑centric tasks and demonstrating that co‑training benefits the task policy even when the controller is omitted at inference.
By Jiatan Huang, Zheyuan Zhang, Ziming Li, Tianyi Ma, Mingchen Li, Yanfang Ye, Chuxu Zhang
OGPO is an Offline Goal‑conditioned Policy Optimization framework designed to improve Vision‑Language‑Action (VLA) models for robotic manipulation. It replaces traditional step‑wise action imitation with adaptive relabeling of trajectories using semantic key states and goal‑conditioned rewards, enabling policies to reach task‑progress states even from off‑demonstration states. Experiments on LIBERO, MetaWorld, LIBERO‑Plus, and the proposed LIBERO‑ReAct benchmark show that OGPO consistently outperforms standard supervised fine‑tuning in terms of robustness and recoverability under distribution shifts.
By Xule Gao, Xi Wang, Zehua Zang, Rui Wang, Changwen Zheng, Chuxiong Sun
SmartSalesBench is a new multi‑turn conversational benchmark designed to evaluate large language model sales agents against buyer personas that are explicitly programmed to resist and challenge the agents. The benchmark features five agent profiles, twelve resistant buyer personas, fifteen smartphone products across six price tiers, and includes 13,680 sessions across four controlled experiments. Evaluation uses nine metrics grouped into primary, secondary, and exploratory tiers, combining automated LLM‑as‑judge scoring with human assessment.
By P N Bhargav Teja, Velpula Lokesh Reddy, Madhu Basumata, Susen Sarkar, Rajasekhara Reddy Duvvuru Muni
The paper introduces two continuous-query paradigms—CQG‑Single and CQG‑AR—to replace the traditional finite-beam, discrete Semantic ID (SID) decoding used in generative recommendation systems. By removing the SID vocabulary and beam expansion, these methods allow every encoded item to be directly scored via dense or approximate-nearest-neighbor retrieval, thereby eliminating the reachability constraints of finite-beam decoding. Experiments show that CQG‑Single outperforms a TIGER-style baseline across all runs while using fewer parameters, and CQG‑AR performs comparably or better on several datasets, suggesting that decoupling output representation, autoregressive modeling, and retrieval budget can improve generative recommendation evaluation.
By Yu Sun, Renkai Xiang, Zuodong Xiang, Junwei Zhou, Yike Zhang, Jianxin Zhang, Zhihao Zhu, Xiaoyu Cao, Hailu Xu
The paper introduces Hindsight Structured Advice Distillation (HSAD), a method that trains a small advisor model to generate structured guidance for a frozen executor language‑model agent. By analyzing executor failures on a training set, a teacher writes advice that would have helped, and the advisor learns to produce this advice from input alone, adding minimal latency. Experiments on the FrontierFinance benchmark show that an 8B advisor boosts the qualification rate of Samaya’s research agent from 52.9% to 56.6%, surpassing the agent’s high‑effort configuration while cutting cost per query by 37%, and the approach also improves performance on WebShop, BIRD text‑to‑SQL, and ExpertLongBench.
By O. Ozan Koyluoglu
KnowForge is a modular pipeline for open‑domain multimodal document question answering that separates retrieval, judging, and reading into distinct agents. It uses a hybrid‑retrieval knowledge‑point index, a Locator agent for searching, a Bridge Judge to enforce a second hop, a Deep Reader that handles text, images, and tables, and a synthesizer with format checking. On the M3DocVQA benchmark it achieves an F1 of 47.5, outperforming prior systems by 11 points, and shows strong gains on MultiHop‑RAG and MMLongBench‑Doc as well.
By Wenlin Lin, Fanzhe Wei
RoleWorld is a symbolic world‑modeling framework that uses reusable prototypes and environment‑specific parameters—called functional role abstractions—to capture how objects behave in interactions. A teacher LLM extracts role prototypes and executable transition rules from interaction trajectories, while an actor LLM employs a skill to apply these roles during decision making. In target environments, the actor refines role assignments and parameter estimates online through active testing, without updating LLM weights, and experiments on the MARS benchmark demonstrate robust generalization across environments with altered object behavior.
By Boyuan Xiao, Pu Wang, Ji Feng, Yao-Xiang Ding
BoundaryEvo introduces a two‑stage data‑model co‑evolution framework for distilling long‑chain reasoning from large teacher models into small language models. First, Capability Priming distills a small set of teacher trajectories to give the student an initial reasoning baseline. Then, Capability‑Boundary Co‑Evolution iteratively selects questions the student has not mastered but the teacher can solve, updating the training data as the student’s capability grows. With only 1,000 examples, BoundaryEvo improves Qwen3.5‑4B accuracy from 62.9 % to 75.3 % on IMO‑AnswerBench and from 26.8 % to 46.0 % on AMO‑Bench, outperforming static and dynamic baselines across 11 benchmarks.
By Jiaming Su, Suorong Yang, Xudong Li, Tengchao Yang, Peng Ye
AlphaNIST demonstrates that a single reinforcement‑learning policy can master 20 diverse assembly tasks on the NIST Assembly Task Board #1 using only a sparse binary reward, without demonstrations, progress shaping, scripted primitives, or per‑task tuning. The policy achieves 98.2% overall training success and can complete ten assembly pairs in a single episode, discovering dexterous, planning‑like behaviors such as selecting future grasps and reorienting parts using surrounding geometry. The work shows that task‑agnostic rewards can enable robots to learn complex manipulation strategies autonomously.
By Octi Zhang, Iretiayo Akinola, Rosario Scalise, Mateo Guaman Castro, Patrick Yin, Tobias Widmer, Miles Macklin, Chris Amevor, Dylan Turpin, Dirk Van Gelder, Jan Carius, Ankur Handa, Gavriel State, Yunrong Guo, Yashraj Narang
PRISM (Population‑Retrieved Inference for Structured Multimodal Prognosis) is a multimodal language model that integrates structured measurements, physiological waveforms, longitudinal history, and population context for clinical prognosis. It uses per‑feature numerical tokenization, recency‑weighted aggregation of ECGs, and a dual‑modality population memory bank, retrieving mean‑pooled embeddings of similar encounters without including individual records in the prompt. On the 1,443‑task MDS‑ED benchmark, PRISM achieves high AUROCs across diagnosis, deterioration, ICU admission, and mortality, outperforming baselines and demonstrating the value of structured multimodal integration and supervised external adaptation.
By Hamza Shafiq, Jialu Tang, Ting Dang, Jun Hu, Aaqib Saeed
The paper introduces MENTOR, a multi‑agent framework that provides on‑policy pedagogical intervention to address zero‑reward collapse in reinforcement learning from verifiable rewards. MENTOR diagnoses failures in real time and generates targeted prompts to guide the learning agent, integrating seamlessly with existing policy optimization pipelines such as GRPO. Experiments on math and code generation tasks show that MENTOR significantly reduces collapse, achieving up to 78.3% relative gains on Llama‑3.2‑3B‑Instruct.
By Guanlan Yue, Hang Yu, Xiaoyong Huang, Ruipeng Li, Wei Zhang, Bingchang Liu, Rui Wang
The paper introduces HOA (Humanize Olympiad Agent), a workflow that orchestrates multiple LLM agent sessions via reviewer feedback, a structured wiki, and a verifier to tackle olympiad problems. Using GPT‑5.6 Sol, HOA achieves perfect scores on the IMO, IPhO, and IOI 2026, near‑full marks on the IChO 2026, and solves all 672 PutnamBench problems, while ranking first on the Lean‑Eval leaderboard. With open models like Kimi‑K3, HOA still attains full scores on major olympiads and a gold‑level IChO score, showing that systematic feedback can close gaps between closed‑source and open‑source agents.
By Zhengyang Zhang, Jing Xiong, Hongzhou Lin, Juihui Chung, Zijian Zhang, Song Han, Sihao Liu, Ligeng Zhu