VidOmni-Bench is a new benchmark for fine‑grained video understanding that asks models to verify whether each event in dense video captions is supported by the video. It contains 500 videos covering five complexity types and durations from 4 seconds to 90 minutes, and uses human‑verified sentence‑level labels to create hard negatives. Experiments show that Video‑LLMs often hallucinate events, struggle to detect incorrect descriptions, and exhibit varying weaknesses depending on video complexity and duration.
By Changbeen Kim, Junwon Chang, Kipyo Kim, Risa Shinoda, Kuniaki Saito, Donghyun Kim
SynthDemo‑RL introduces a teacher‑student framework that uses an automated teacher to generate successful manipulation trajectories from simulator‑privileged state, which are then distilled into a Vision‑Language‑Action (VLA) student via supervised fine‑tuning. The student is further refined with PPO using binary task‑success rewards. On the LIBERO‑PRO benchmark, SynthDemo‑RL rescues all 27 previously unsolvable tasks and achieves near‑perfect success rates, matching performance that would otherwise require human demonstrations.
By Hiroaki Kingetsu, Hiroaki Kurihara, Kaoru Yokoo, Kenji Fukumizu, Manohar Kaul
CIBuzzBench is a new benchmark that tests cross‑lingual understanding of Chinese internet buzzwords by providing 3,001 buzzwords with English explanations, equivalents, category labels, and harmfulness annotations. The benchmark defines three tasks—Meaning Explanation, Cross‑lingual Equivalent Matching, and Culturally Grounded Harmfulness Detection—to evaluate how well large language models translate and interpret these culturally nuanced terms. Experiments show that current state‑of‑the‑art LLMs still struggle with fine‑grained non‑literal meanings, robust matching, and accurate harmfulness detection across languages.
By Yifan Wang, Junyu Lu, Qifan Wang, Shun Zhang, Chaozhuo Li, Jiahao Liu, Zhijun Cao, Lingbin Bu, Fanliang Bu
arXiv:2609.21942v1 Announce Type: cross
Abstract: A robot that fails at a task faces the first decision in corrective dialogue: act on its own diagnosis, consult another onboard sensor, or interrupt...
By Eshika Pathak, Leela Krishna
CRYSTAL is a diagnostic benchmark comprising 6,372 multimodal reasoning instances that assess models through verifiable intermediate steps. It introduces two metrics—Match F1 and Ordered Match F1—to evaluate step-level precision, recall, and order. The benchmark, built via a Delphi-inspired pipeline with four independent MLLMs, reveals systematic failures in current models, such as cherry‑picking and disordered reasoning, and proposes a Causal Process Reward and CPR‑Curriculum to improve reasoning performance.
By Wayner Barrios, SouYoung Jin
The paper introduces Generative Executable Algorithm Knowledge Graphs (GEAKG), a graph-based representation that stores procedural knowledge for algorithm design as typed nodes with validated operators, edges encoding admissible compositions, and learned edge weights capturing effective sequences. GEAKG can be instantiated across different domains by altering only a role ontology and binding, enabling transfer of knowledge. Experiments show that within neural‑architecture‑search families, learned snapshots transfer across many dataset pairs, while across combinatorial domains only the ontology‑constrained executable structure transfers, providing significant savings in expensive target‑side searches such as large scheduling instances.
By Camilo Chac\'on Sartori, Jos\'e H. Garc\'ia, Andrei Voicu Tomut, Christian Blum
BoostAPR is a three-stage framework that improves automated program repair by using execution-grounded reinforcement learning with dual reward models. The approach first fine‑tunes a model on execution‑verified demonstrations, then trains a sequence‑level assessor and a line‑level credit allocator from execution outcomes, and finally applies PPO optimization where the line‑level model redistributes rewards to critical edit regions. Evaluated on SWE‑Gym and four benchmarks, BoostAPR achieves significant gains, including 40.7% on SWE‑bench Verified and 95.0% on QuixBugs, demonstrating strong cross‑language generalization.
By Yuanhao Li, Hongbo Wang, Xiaotang Shang, Xunzhu Tang, Yiming Cao, Xuhong Chen
The paper introduces Balance of Benchmarks (BoB), a framework that improves task-conditioned model comparison by weighting benchmark evidence based on semantic density, equating scores across varying difficulty levels, and pooling task-relevant residuals. BoB retains all eligible benchmark data while adjusting its influence, outperforming uniform averaging on the WildScores dataset with higher Spearman correlation, lower MAE, and better shortlist hit rates. The method also reduces ranking instability when benchmarks are repeated or paraphrased, and lowers retrospective regret in model selection.
By Jhen-Ke Lin, Hong-Yun Lin
The paper presents a comparative study of motion planning methods from major autonomous driving benchmarks—CARLA, nuPlan, and the Waymo Open Dataset—using CARLA Leaderboard v2.1 as a unified evaluation platform. Eight representative planners (TF++, InterFuser, TCP, PDM‑Lite, MTR+MPC, CaRL, PlanT 2.0, Diffusion planner) are evaluated to highlight their strengths, weaknesses, prevailing trends, and common challenges in motion planning research.
By Merve Atasever, Alfredo Reina Corona, Zhuochen Liu, Qingpei Li, Akshay Hitendra Shah, Hans Walker, Jyotirmoy V. Deshmukh, Rahul Jain
The paper introduces a chemical language foundation model that uses a SMILES‑based polymer graph representation (CPG) to encode polymer architecture and connectivity, addressing gaps in existing line notations. The model achieves strong performance across 30 polymer property benchmarks and demonstrates robustness to structural representation perturbations, with even chemically invalid SMILES sometimes matching state‑of‑the‑art results. Control experiments and attention analyses confirm that CPG offers meaningful advantages while highlighting the model’s ability to interpolate SMILES sequence space in a way that aligns loosely with chemical and architectural space.
By Nathaniel H. Park, Eduardo Soares, Victor Y. Shirasuna, Tiffany J. Callahan, Sara Capponi, Emilio Vital Brazil
BEAT-Net is a supervised biomimetic framework for ECG diagnosis that incorporates QRS-centered tokenization and a hierarchical architecture mirroring a cardiologist’s workflow. It processes heartbeat sequences through morphological, spatial, temporal, and transformer-based stages, achieving an AUC of 0.924 on large benchmarks while using only 0.7 million parameters. The model outperforms the 39.5‑million‑parameter HeartLang foundation model on morphological form classification and demonstrates superior cross‑dataset generalization with only 35% of the training data.
By Runze Ma, Haonan Lyu, Shunbo Jia, Qiang Yang, Muzi Xu, Jiaqi Zhang, Zihe Luo, Caizhi Liao
The paper introduces a multilingual story moral generation task to evaluate cultural alignment in large language models. Using a dataset of human-written story morals from 14 language‑culture pairs, the authors compare model outputs to human interpretations through semantic similarity, a preference survey, and value categorization. They find that advanced models like GPT‑4o and Gemini produce morally similar and preferred responses but show less cross‑linguistic variation, focusing on a narrower set of shared values, indicating a limitation in capturing the diversity of human narrative understanding.
By Sophie Wu, Andrew Piper
Rhamba is a region‑aware pretraining framework for resting‑state fMRI that combines anatomically guided masking with hybrid Attention‑Mamba architectures. The study pretrained models on the ABIDE dataset using three masking strategies (Any, Majority, Pure) and evaluated four architectural variants, finding that the Mamba‑Attention (MA) hybrid achieved the best average AUROC on downstream schizophrenia and ADHD classification tasks. Explainable AI via Integrated Gradients highlighted that performance depends on the interaction between masking strategy and architecture rather than a single dominant configuration.
By Pankaj Pandey, Ruthwik Reddy Doodipala, Pratheek Eranki, Carolina Torres-Rojas, Manob Jyoti Saikia, Ranganatha Sitaram
arXiv:2605.06445v2 Announce Type: replace-cross
Abstract: Large Language Model (LLM) agents demonstrate strong performance in autonomous code generation under loose specifications. However, productio...
By Francesco Dente, Dario Satriani, Paolo Papotti
LiteMedCoT-VL is a parameter‑efficient pipeline that transfers chain‑of‑thought reasoning from a 235B teacher model to a 2B student model using LoRA fine‑tuning on explanation‑enriched data. The approach enables a compact vision‑language model to perform medical visual question answering without relying on image captions, achieving 64.9% accuracy on the PMC‑VQA benchmark—an 11‑point improvement over the zero‑shot Qwen3‑VL‑4B baseline. Visual grounding analysis confirms that the model bases its predictions on image content rather than textual priors.
By Runze Ma, Shunbo Jia, Haonan Lyu, Guo Liu, Caizhi Liao
The paper introduces a training‑free proactive defense for detecting partial deepfake speech by using self‑embedding steganography. It embeds a compressed version of the clean audio within itself, allowing post‑hoc extraction of reference content and enabling detection of spoofed segments via codec‑based restoration. Experiments on a benchmark dataset show that this method complements passive detectors and operates without any training, offering a robust, data‑efficient alternative for partial deepfake detection.
By Yigitcan \"Ozer, Zhe Zhang, Wanying Ge, Xin Wang, Junichi Yamagishi
CompAdapt is a physics-consistent text-to-video generation framework that extends diffusion-based models to handle composite physical behaviors such as coupled motions, multi-stage transitions, and multi-object collisions. It translates natural language prompts into structured physical semantics, enabling end-to-end specification of motion types, temporal relations, and initial parameters. The system introduces dynamics-aware prior matching for one-shot adaptation to new physical environments and a physics-aware latent feature fusion module to enhance visual fidelity during fast, complex motion, outperforming existing physics-constrained baselines on physics-focused T2V benchmarks.
By Haoran Qin (Harbin Institute of Technology, China), Renlong Wu (Harbin Institute of Technology, China), Tianyu Huang (Harbin Institute of Technology, China), Yukang Ding (Taobao, Alibaba Group, China), Hui Li (Harbin Institute of Technology, China), Wangmeng Zuo (Harbin Institute of Technology, China)
The paper introduces an end‑to‑end model‑based reinforcement learning algorithm that synthesises policies satisfying Linear Temporal Logic (LTL) specifications in unknown environments. It synchronises a Limit‑Deterministic Büchi Automaton (LDBA) with a Bayes‑Adaptive Markov Decision Process (BAMDP) and proposes a novel Bayes‑Adaptive Monte‑Carlo Planning (BAMCP) method for approximate Bayes‑optimal strategy synthesis. Experiments on finite and infinite‑horizon tasks show improved property satisfaction and sample efficiency compared to model‑free baselines, and ablation studies confirm the advantage of the new BAMCP over classical variants, including reduced task violations in cautious RL settings.
By Jonathan Hau, Alessandro Abate
DisasterInsight is a building‑centric benchmark designed to evaluate vision‑language models (VLMs) for disaster response. Built on the xBD satellite dataset, it adds OpenStreetMap‑derived functional labels to 134,108 building instances and offers 15 task types, including instance assessment, scene counting, multi‑instance reasoning, and structured report generation. Experiments show that VLMs excel at visible damage detection but struggle with building function, multi‑instance reasoning, counting, and grounded reporting, and instruction tuning only partially mitigates these gaps.
By Sara Tehrani, Yonghao Xu, Leif Haglund, Amanda Berg, Gulnaz Zhambulova, Michael Felsberg
arXiv:2609.21190v1 Announce Type: cross
Abstract: Ensuring the correctness of LLM-generated code is a core challenge for modern software engineering. Benchmarks for agentic code generation check corr...
By George Ma, Benjamin Mikek, Haoyu Li, Ferhat Erata, Yuhao Zhang, Zeren Shui, Behrooz Omidvar Tehrani, Jun Huan, Murali Krishna Ramanathan, Somayeh Sojoudi, Hao Zhou, Anoop Deoras