Proposes “Roach”, a reinforcement learning expert surpassing Autopilot in CARLA, providing rich supervision signals for end-to-end imitation learning.

Zhejun Zhang, Alexander Liniger, Dengxin Dai et al.

Moser Flow (MF), a divergence-based continuous normalizing flow, avoids ODE solving, enabling efficient sampling and density estimation on complex manifolds.

Noam Rozen, Aditya Grover, Maximilian Nickel et al.

Proposes a boundary-focused 3D point cloud reconstruction framework using differentiable projection, outperforming existing methods.

Guiju Ping, Mahdi Abolfazli Esfahani, Han Wang

Defines foundation models, analyzes emergent capabilities and risks, emphasizes interdisciplinary research importance.

Rishi Bommasani, Drew A. Hudson, Ehsan Adeli et al.

V-DeepONet combines physics and minimal data to swiftly predict crack paths in brittle materials.

Somdatta Goswami, Minglang Yin, Yue Yu et al.

Survey of safe learning methods in robotics, integrating model learning, robust control, and formal safety verification, emphasizing algorithmic frameworks and real-world applications.

Lukas Brunke, Melissa Greeff, Adam W. Hall et al.

MUSIQ employs multi-scale Transformer processing native images, surpassing CNN fixed-size constraints, achieving state-of-the-art results.

Junjie Ke, Qifei Wang, Yilin Wang et al.

Proposed Mobile-Former combines MobileNet and Transformer with few learnable tokens, achieving high accuracy at low FLOPs.

Yinpeng Chen, Xiyang Dai, Dongdong Chen et al.

Introduces DexMV, combining video-based 3D pose estimation and imitation transfer, boosting dexterous manipulation success rates by over 85%.

Yuzhe Qin, Yueh-Hua Wu, Shaowei Liu et al.

Proposes MIMO-UNet, integrating multi-scale inputs and multi-output decoding, achieving faster and more accurate single-image deblurring.

Sung-Jin Cho, Seo-Won Ji, Jun-Pyo Hong et al.

A Loihi event-driven Hough-SNN reaches 20 kHz vision updates, tracks 1200°/s motion, and adapts online to a 125 g UAV disturbance.

Antonio Vitale, Alpha Renner, Celine Nauer et al.

This study evaluates six offline algorithms on robot manipulation tasks, highlighting data quality and model design importance.

Ajay Mandlekar, Danfei Xu, Josiah Wong et al.

Physics-based noise model for extreme low-light imaging enables realistic synthetic data for deep denoising, outperforming real-data training.

Kaixuan Wei, Ying Fu, Yinqiang Zheng et al.

ObsNet with Local Adversarial Attacks achieves fast and accurate OOD detection in semantic segmentation.

Victor Besnier, Andrei Bursuc, David Picard et al.

SDEdit leverages pre-trained diffusion models with reverse SDE denoising to enable task-agnostic image synthesis and editing, outperforming GANs with up to 98% realism improvement.

Chenlin Meng, Yutong He, Yang Song et al.

Proposed a hybrid reasoning system achieving 56.8% accuracy on LSAT.

Siyuan Wang, Zhongkun Liu, Wanjun Zhong et al.

Perceiver IO: a universal, linear-scaling architecture for arbitrary inputs/outputs, outperforming Transformers in multi-task/multi-modal tasks.

Andrew Jaegle, Sebastian Borgeaud, Jean-Baptiste Alayrac et al.

ManiSkill benchmark introduces diverse 3D assets and large-scale demonstrations, advancing generalizable robotic manipulation skills.

Tongzhou Mu, Zhan Ling, Fanbo Xiang et al.

UIBert trains on unlabeled data with five pre-training tasks, improving UI understanding accuracy by up to 9.26%.

Chongyang Bai, Xiaoxue Zang, Ying Xu et al.

A systematic survey of prompting methods in NLP, detailing models, prompt design, answer engineering, and experimental validation with specific datasets.

Pengfei Liu, Weizhe Yuan, Jinlan Fu et al.