Mini Amusement Parks (MAPs) is a testbed for modeling business decisions, with experts outperforming AI by 11.4x.

Stéphane Aroca-Ouellette, Ian Berlot-Attwell, Panagiotis Lymperopoulos et al.

VisPlay uses self-evolving RL to improve vision-language models' reasoning via unlabeled image data.

Yicheng He, Chengsong Huang, Zongxia Li et al.

Proposed DRP framework enhances LLM personalization via inference scaling, achieving a 23% BLEU improvement.

Suyu Chen, Yimeng Bai, Yulong Huang et al.

Poetic style as a universal single-turn jailbreak, achieving 62% attack success across 25 LLMs, with high cross-domain transferability.

Piercosma Bisconti, Matteo Prandi, Federico Pierucci et al.

EntroPIC employs proportional-integral control to stabilize entropy, enhancing exploration in large language model training.

Kai Yang, Xin Xu, Yangkun Chen et al.

SciEGQA introduces semantic-region grounding and Grounding–Crop–then–Answer, with 1,623 human QA pairs and 30K+ training pairs.

Wenhan Yu, Zhaoxi Zhang, Wang Chen et al.

HAPPO achieves over 95% load restoration in large-scale distribution systems using heterogeneous multi-agent PPO with centralized critic.

Parya Dolatyabi, Ali Farajzadeh Bavil, Mahdi Khodayar

Agent-R1 introduces step-level trajectory abstraction and flexible context management, enabling multi-turn reinforcement learning with diverse optimization strategies, achieving state-of-the-art results.

Mingyue Cheng, Shuo Yu, Daoyu Wang et al.

ARC-Chapter leverages large-scale multimodal data and hierarchical annotations, achieving 14% F1 and 11.3% SODA improvements on long video chaptering.

Junfu Pu, Teng Wang, Yixiao Ge et al.

Introduced soft-label training, reducing KL divergence by 32% on datasets like ChaosNLI, enhancing model uncertainty expression.

Agamdeep Singh, Ashish Tiwari, Hosein Hasanbeig et al.

PhysX-Anything uses VLM and a novel 3D tokenization to generate high-quality, physically grounded assets from a single image, enabling direct simulation deployment.

Ziang Cao, Fangzhou Hong, Zhaoxi Chen et al.

M-GRPO extends Group Relative Policy Optimization for hierarchical multi-agent systems, aligning heterogeneous trajectories to improve reasoning performance.

Haoyang Hong, Jiajun Yin, Yuan Wang et al.

This study introduces SpatialSky-Bench and Sky-VLM, achieving SOTA in 13 UAV spatial reasoning tasks with 53.3 average score, surpassing baselines by over 130%.

Lingfeng Zhang, Yuchen Zhang, Hongsheng Li et al.

GeoX-Bench benchmarks large multimodal models for cross-view geo-localization and pose estimation, showing strong localization but challenges in pose accuracy.

Yushuo Zheng, Jiangyong Ying, Huiyu Duan et al.

DGS-Net uses gradient decomposition and distillation to improve CLIP fine-tuning for AI image detection, achieving a 6.6% accuracy boost.

Jiazhen Yan, Ziqiang Li, Fan Wang et al.

Spark-Prover-X1 enhances formal theorem proving via diverse data training, solving 27 problems on PutnamBench.

Xinyuan Zhou, Yi Lei, Xiaoyu Zhou et al.

Uni-Inter synthesizes 3D human motion across diverse interaction contexts using a Unified Interactive Volume (UIV).

Sheng Liu, Yuanzhi Liang, Jiepeng Wang et al.

Scaled GenSelect and LLM-as-a-Judge to millions of tokens, enhancing math proof verification.

Sadegh Mahdavi, Branislav Kisacanin, Shubham Toshniwal et al.

GEM uses entropy-guided generative preference modeling for few-shot LLM alignment, improving preference prediction by 5-10% and task performance significantly.

Yiyang Zhao, Huiyu Bai, Xuejiao Zhao

Empirical analysis of 2,303 agent context files reveals their structure, maintenance, and content biases, highlighting insufficient emphasis on security and performance.

Worawalan Chatlatanagulchai, Hao Li, Yutaro Kashiwa et al.