SceneReVis is a vision‑grounded, self‑reflective framework for 3D indoor scene synthesis that uses an iterative diagnose‑and‑act loop to detect and fix spatial conflicts such as collisions. It introduces SceneChain‑12k, a large dataset of causal construction trajectories generated via a reverse engineering pipeline, and employs a two‑stage training recipe that moves from supervised fine‑tuning to agentic reinforcement learning, turning the model into an active spatial planner. Experiments show that SceneReVis achieves state‑of‑the‑art performance in high‑fidelity generation and goal‑oriented optimization, with better transfer to underrepresented room categories and potential for out‑of‑distribution room types.
By Yang Zhao, Shizhao Sun, Meisheng Zhang, Yingdong Shi, Xubo Yang, Jiang Bian
The paper introduces SADGE, a quantitative similarity metric that predicts how well synthetic image datasets will perform on real-world computer vision tasks without requiring downstream model training. SADGE combines appearance and geometric similarity measures through a constrained bilinear interaction, outperforming appearance-only and geometry-only baselines across 79,000 image pairs in five synthetic-to-real benchmark families. The study demonstrates that the non‑linear interplay between appearance and geometry, rather than either alone, best predicts downstream performance in tasks such as object detection, semantic segmentation, and pose estimation.
By Patryk Bartkowiak, Dominik Michels, Soren Pirk, Wojtek Palubicki
The paper proposes InterpTok, a method that fine‑tunes the MAETok image tokenizer to better handle codes that lie between training samples. By training the decoder on mixtures of image codes and using a discriminator plus a mixing‑ratio loss, InterpTok preserves reconstruction quality while dramatically improving interpolation and generation metrics. The approach yields sharper, more natural image compositions and significantly lowers FID scores for diffusion models trained on the new tokenizer.
By Ole Bei{\ss}wenger, Jan-Niklas Dihlmann, Hendrik Lensch
The paper introduces TPMDM, a diffusion model that decouples and aligns heterogeneous geometric constraints—user-defined global trajectories and local image-derived poses—through a parallel control strategy. Building on TPMDM, the IKMo framework integrates a VLM-powered multi-agent system to translate user-friendly inputs into explicit motion constraints, enabling precise animation from unaligned, in-the-wild 2D reference images. Experiments on HumanML3D and KIT-ML show that IKMo achieves state‑of‑the‑art performance for motion generation under mixed trajectory and keyframe pose constraints.
By Yang Zhao, Yan Zhang, Zixuan Guo, Hangyu Zhou, Haotian Mao, Hui Wang, Xubo Yang
StepSafe introduces a training framework called ASPO that focuses on step‑level timely intervention in agentic systems. It aligns decisions with their respective prefixes, rewards actions that fit the current stage, and uses adaptive feedback self‑distillation to credit reasoning. The resulting StepSafe guardrail achieves state‑of‑the‑art performance on five agent‑safety benchmarks, improving the safety‑utility Pareto curve with a Youden’s J of 53.0 for a 4B model.
By Siyi Gu, Kunhan Gao, Chao Peter Yang, Sukanta Ganguly, Rex Ying, Arman Cohan
The paper introduces a three‑level shortcut‑auditing framework to assess AI guard reliability in agent‑safety evaluations. It demonstrates that shortcut cues—such as shorter shell commands or exact‑tool lookups—can predict trajectory labels and influence guard decisions, leading to significant changes in measured performance when benchmark representations are altered. Experiments on LinuxArena, AgentDojo‑Traj, and AgentHarm‑Traj datasets reveal that guard accuracy can drop dramatically under domain holdouts and that editing safe‑row markers or tool aliases materially affects false‑positive rates.
"whyItMatters":"The study shows that benchmark design can directly alter guard decisions and evaluation outcomes, underscoring the need for careful evaluation protocols to ensure reliable safety assessments."
By Hasan Demirkiran, Jens Ernstberger
The paper proposes a hierarchical conditioning approach for joint Acoustic Scene Classification and Audio Tagging, predicting scene categories directly from intermediate event probabilities. Extensive experiments on three backbones and varied loss‑weighting settings show that this method matches or surpasses traditional multi‑head models while enabling synchronized dual‑task learning for certain beta values. An interpretability framework combining information bottleneck analysis, manifold geometry, and Formal Concept Analysis on TF‑IDF event representations demonstrates that hierarchical conditioning reduces conditional entropy, increases mutual information, and improves latent cluster separability, yielding a more structured and semantically faithful feature space.
By Claudia Montero-Ram\'irez, Alba Mart\'inez-Serrano, Carmen Pel\'aez-Moreno
The paper introduces AI Consilium, a benchmark for evaluating large language models (LLMs) on providing clinically grounded, coherent, actionable, and safe second opinions in multidisciplinary settings. Using 300 NEJM–MedQA cases, the study compares eight single-model pipelines that generate role-specific opinions and interactive-consilium reports under free-form and structured conditions. Results show that structured final reports generally score higher, especially in management, and that GPT‑5.6‑Sol outperforms other models across all evaluated components, demonstrating the benchmark’s ability to differentiate model capabilities and the impact of response structure.
By Alexey A. Bokov, Ivan I. Panin, Nikolay B. Nechaev, Mikhail Yu. Valkov, Valeria Yu. Chernina, Natalia Gombolevskaya, Ayyaz Hussain, Vladimir M. Agafonov, Victor A. Gombolevskiy, Aleksandr V. Nesterov, Iaroslav R. Bespalov, Ivan V. Oseledets, Dmitry V. Dylov
The paper introduces an Availability Contract to address timing mismatches in financial NLP benchmarks, requiring explicit declaration of input boundaries, evidence of availability, outcome start, and supported claims. Applying this framework to 552 earnings calls, the authors show that shifting from a call‑start proxy to a completion proxy changes return signs in 9.4% of calls and alters quintile rankings in 28.3%. In a volatility benchmark, moving target windows forward by one session increases mean squared error by 18.1‑21.4% while leaving point rankings unchanged, illustrating that labels and losses can be sensitive to the declared outcome clock.
By Xiaoyu Ma, Kelvin J. L. Koa, Yuncong Liu, Ke-Wei Huang
PremiseSQL introduces a benchmark of 1,800 flawed natural language to SQL (NL2SQL) queries, grouped into four categories of premise failures, alongside their original valid counterparts. The study evaluates large language models’ ability to critique flawed premises and abstain from generating misleading SQL, using an evaluation framework across three prompting settings. Results show that current models tend to trust premises and rarely critique them, while stronger critique-oriented prompting improves detection but also raises false invalid judgments and reduces SQL generation for valid queries.
By Shuo Lin, Jinzhe Li, Chenlu Guo, Yanhan Li, Yuan Wu, Yi Chang
NeoHorse-1 is a family of agent-native models that pursue recursive self‑improvement through agentic post‑training. The system uses a heterogeneous model pool with intelligent routing to record capability demands, service tiers, and interactions, converting these records into training examples that preserve reasoning, tool calls, and harness context. By structuring supervised fine‑tuning into a three‑stage curriculum and employing routing‑guided on‑policy distillation, NeoHorse-1 closes an evaluation‑selection‑update loop, raising macro‑average performance from 58.94 to 64.87 on 4B models and from 65.60 to 69.04 on 9B models across eleven benchmarks, while also producing a prefill‑only variant that outputs structured decisions and probabilities for agent workflows.
By NeoHorse Team, Guoliang Cao, Guohao Dai, Tianyu Guo, Kai Han, Hailin Hu, Zihan Jiang, Xiang Kuang, Boxun Li, Yulong Li, Zehua Pei, Yuchuan Tian, Jiamin Wang, Yu Wang, Yunhe Wang, Yihong Wu, Haiyang Xu, Shuo Zhang, Hang Zhou, Siyang Cheng, Jiayu Fan, Wei He, Qingrui Jiao, Hongguang Li, Zhiyuan Li, Runke Liu, Xi Liu, Xinchen Liu, Sinno Jialin Pan, Yi Ren, Liuyang Song, Chenyu Wang, Bei Yu, Quanlu Zhang, Xiangyu Zhang, Mengyu Zheng, Yingjie Zong
SUPRA-Tex is a new method for single‑image 3D generation that dresses a 3D mesh with relightable materials while preserving the visual details of a reference image. It decouples visible‑material estimation, surface alignment, and observation‑guided 3D completion, allowing a 3D generator trained on limited data to focus on preserving visible evidence and completing unobserved surfaces. The approach uses an MLLM‑based image editor to predict lighting‑separated PBR maps, projects them onto the visible surface, encodes them with a pretrained material VAE, and employs an interleaved flow transformer to generate complete material latents, achieving state‑of‑the‑art performance on geometry‑conditioned texturing and end‑to‑end asset generation.
By Chen Yang, Jiemin Fang, Fanpeng Meng, Zihao Huang, Jichen Pang, Guanjun Wu, Zhikuan Bao, Qi Tian
LoopWAM is a closed‑loop world‑action model that learns from the full spectrum of real‑world rollouts—successes, suboptimal actions, and failures—by combining behavior learning with action‑conditioned world modeling. It uses source‑dependent noise sampling, quality steering, and quality‑weighted supervision to retain video supervision across all quality levels while regulating action imitation. On several robotics benchmarks, including RoboChallenge Table30‑V2, RoboTwin 2.0, RoboCasa365, and a shirt‑folding task, LoopWAM achieves significant performance gains after a few deployment‑and‑training rounds, outperforming comparable methods.
By Wenxuan Guo, Yingping Liang, Zhaoguo Wang, Ziyuan Li, Angyuan Ma, Zhenyu Wu, Chengkun Wang, Bingyao Yu, Ze Chen, Borui Zhang, Xiuwei Xu, Meng Zhang, Erjin Zhou, Jiwen Lu
BD-HazardVLM is a diagnostic benchmark of 500 Bangladesh road-scene images designed to probe vision‑language models’ defensive‑driving reasoning in weakly structured Global South traffic. The benchmark includes annotations for hazard presence, category, risk timing, severity, defensive actions, visible evidence, and bias traps, and evaluates models such as Qwen2.5‑VL‑7B, InternVL3‑8B, and MiniCPM‑V‑4.6. Results show high binary hazard‑presence accuracy but poor fine‑grained reasoning, with category accuracy below 40 %, risk‑timing accuracy under 13 %, and defensive‑action matching under 2 %.
whyItMatters":"The study demonstrates that coarse hazard recognition does not guarantee reliable fine‑grained defensive‑driving reasoning, highlighting the need for geographically diverse safety evaluation of vision‑language models."
By Abdullah Al Maruf, Md. Sajedul Islam, Tanmoy Mridha, Irfan Hossain Bhuiyan
The paper introduces a large corpus of about 160,000 Brazilian Portuguese financial news articles to support Natural Language Processing research in this domain. It presents initial benchmarks for a weakly supervised multi‑class classification task using semantic similarity, evaluating cosine similarity, K‑Nearest Neighbors (KNN), and FAISS‑based nearest‑neighbor search with multilingual embeddings. Results show that KNN performs best on manually labeled data (47.3% accuracy, 31.9% F1), while LLM‑annotated data yields similar accuracy but much lower F1 scores due to label imbalance and noise.
By Leomax Bandeira, Gabriel Aguiar, Dalton C\'ezane Gomes Valadares, Luiz Pereira, Helder Fernando de Ara\'ujo Oliveira, Gustavo Xavier, Andson M. Balieiro
Graph Additive Models (AMIGA) address three key limitations of Graph Neural Additive Networks (GNANs): they prevent irrelevant nodes from inheriting spurious relevance, represent topology via mined motifs instead of only shortest‑path distances, and separate feature, structure, and interaction effects into distinct additive channels. AMIGA’s decomposition is centered around a graph‑balanced training baseline, ensuring a unique and well‑defined representation for each node. Across ten graph classification and regression benchmarks, AMIGA outperforms GNAN on most datasets and narrows the performance gap with standard black‑box GNNs, while providing node‑local explanations that exactly reconstruct the model output.
By Clara Alvarez-Rodriguez, Steve Azzolin, Vincenzo Marco De Luca, Filippo Maria Bianchi, Andrea Passerini, Antonio Longa
The paper introduces a declarative framework for anticipatory reasoning about interactive motion in dynamic environments. It combines an ASP model of trajectory dynamics, semantically-guided prediction of branching futures, and an energy-based optimization to evaluate and rank anticipations. The approach is evaluated on the Argoverse 2 benchmark, providing explicit, inspectable event-level hypotheses for each anticipated trajectory.
By Julius Monsen, Jakob Suchan, Mehul Bhatt
Kolab is a new dataset comprising over 115 hours of screen recordings from 50 professionals across 40 occupations who collaborate with AI agents on realistic, long‑horizon tasks. The dataset includes participants’ judgments of collaboration quality and output, revealing varied coordination patterns and the need for agents to anticipate how experts will use or modify their contributions. Analysis shows current state‑of‑the‑art models overestimate collaboration success and struggle to differentiate high‑ from low‑quality interactions, indicating a gap in understanding long‑horizon human‑agent teamwork.
By Shirley Anugrah Hayati, Patricia Anugrah Setiani, Ruizi Wang, Zhouyi Li, Khanh Chi Le, Young-Jun Lee, Dongyeop Kang
The paper introduces Condition-Contrastive Extrapolated On-Policy Distillation (CCE-OPD), a method that refines on-policy distillation of language models by making teacher supervision sensitive to question conditions. CCE-OPD uses a fixed-prefix comparison and a bounded log-probability contrast to reweight teacher outputs, reducing invalid continuation preferences from 27.0% to 16.0% on diagnostic reasoning states. Experiments on six mathematical benchmarks show that CCE-OPD improves average performance over standard OPD by 1.60–1.95 percentage points when distilling Qwen3-8B into smaller students, and achieves the best results on four transfer benchmarks.
By Wenxuan Luo, Chenxi Li, Haoyue Liu, Yongxin Guo, Xiaoying Tang
The paper introduces CA‑MAMBA, a constrained multi‑agent reinforcement learning method that predicts cost within a world model, enabling most policy updates to occur in imagination. Compared to a strong model‑free baseline, CA‑MAMBA requires 10–20 times fewer interactions on the median SMAC configuration and achieves 2.8–12 times higher MAMuJoCo rewards for the same budget. In controlled experiments, aligning the model’s cost estimates with a cost‑blind learner reduces that learner’s cost by more than half and satisfies a strict continuous‑control limit, indicating that accurate cost estimation is not the limiting factor.
By Raz Shmueli, Shahaf S. Shperberg