Introduces MMRB2 benchmark, evaluating omni reward models across four tasks; Gemini 3 Pro achieves 75-80% accuracy, surpassing GPT-4o's 59%.

Yushi Hu, Reyhane Askari-Hemmat, Melissa Hall et al.

Kling-Omni is a unified multimodal video synthesis framework integrating instruction understanding, editing, and reasoning, achieving high fidelity and efficiency.

Kling Team, Jialu Chen, Yuanzheng Ci et al.

VLA models integrate vision, language, and action to enhance interpretability and generalizability in autonomous driving.

Tianshuai Hu, Xiaolu Liu, Song Wang et al.

Proves that conditioning e-variables on sufficient statistics monotonically increases expected utility, extending Rao-Blackwell theorem to e-statistics.

Dante de Roos, Ben Chugg, Peter Grünwald et al.

Proposes TODSynth framework with MM-DiT and CRFM, significantly improving synthetic data quality for remote sensing semantic segmentation.

Yunkai Yang, Yudong Zhang, Kunquan Zhang et al.

This study introduces JustRL, a simple fixed-hyperparameter RL training method for 1.5B models, surpassing complex approaches with more stability and less compute.

Bingxiang He, Zekai Qu, Zeyuan Liu et al.

The paper establishes non-asymptotic global convergence rates for PPO-Clip with f-divergence regularization, deriving linear convergence for forward KL and local convergence for reverse KL.

Yin Liu, Qiming Dai, Junyu Zhang et al.

N3D-VLM integrates native 3D perception with spatial reasoning, achieving state-of-the-art 3D grounding and inference.

Yuxin Wang, Lei Ke, Boqiang Zhang et al.

This survey systematically reviews spatio-temporal knowledge graph models, analyzing their origins, modeling dimensions, and challenges.

Philipp Plamper, Hanna Köpcke, Anika Groß

Defines Scientific General Intelligence (SGI) based on Practical Inquiry Model, evaluates models via SGI-Bench across four core tasks with over 1,000 samples.

Wanghan Xu, Yuhao Zhou, Yifan Zhou et al.

This study benchmarks 6 SpeechLLMs across 16 datasets, revealing that while cascaded systems remain reliable, recent SpeechLLMs can match or outperform them, emphasizing LLM integration.

Sara Papi, Javier Garcia Gilabert, Zachary Hopton et al.

OS-Oracle framework achieves state-of-the-art cross-platform GUI critic model performance with a 310k sample dataset.

Zhenyu Wu, Jingjing Xie, Zehao Li et al.

BayesSum uses Gaussian processes to efficiently estimate expectations in discrete spaces, reducing sample requirements by over 50%.

Sophia Seulkee Kang, François-Xavier Briol, Toni Karvonen et al.

VoP method leverages web-scale knowledge, boosting ML models' success rate in city navigation from 20-30% to over 60%.

Dwip Dalal, Utkarsh Mishra, Narendra Ahuja et al.

Mimic-video combines pretrained video models with inverse dynamics decoders, achieving 10x sample efficiency and 2x faster convergence in robotic control tasks.

Jonas Pai, Liam Achenbach, Victoriano Montesinos et al.

Proposes 'Off-The-Grid' architecture for sub-pixel Gaussian primitive detection, significantly improving 3D scene reconstruction quality and efficiency.

Arthur Moreau, Richard Shaw, Michal Nazarczuk et al.

SoliReward enhances video generation reward models' robustness using single-item binary annotations and Hierarchical Progressive Query Attention.

Jiesong Lian, Ruizhe Zhong, Zixiang Zhou et al.

The study finds semantically equivalent code transformations can reduce LLM membership inference success, with RenameVariable rule lowering success by 10.19%.

Hua Yang, Alejandro Velasco, Thanh Le-Cong et al.

EVICPRESS jointly optimizes KV-cache compression and eviction, reducing inference latency while maintaining quality.

Shaoting Feng, Yuhan Liu, Hanchen Li et al.

Proposes On-policy Expert Corrections (OEC) to mitigate covariate shift in multi-turn LM training, improving software engineering task success rates by 13-14%.

Niklas Lauffer, Xiang Deng, Srivatsa Kundurthy et al.