Proposes sublinear algorithms for Wasserstein and TV distances from streaming data, enabling scalable fairness and privacy auditing.

Debabrota Basu, Debarshi Chanda

RoboCopilot system enhances robot bimanual manipulation skills via interactive imitation learning, showing performance improvement in experiments.

Philipp Wu, Yide Shentu, Qiayuan Liao et al.

SANDRO combines IRLS and splitting strategy, achieving 20% higher success in point cloud registration under high outlier rates, outperforming state-of-the-art methods.

Michael Adlerstein, João Carlos Virgolino Soares, Angelo Bratta et al.

DreamRelation customizes videos with relation decoupling and dynamics enhancement, achieving a 30% accuracy improvement.

Yujie Wei, Shiwei Zhang, Hangjie Yuan et al.

LMM-R1 employs a two-stage rule-based RL, boosting 3B multimodal models' reasoning by 4.83% on average, enabling effective transfer from text to vision tasks.

Yingzhe Peng, Gongrui Zhang, Miaosen Zhang et al.

PointVLA selectively injects 3D point cloud features into frozen pre-trained VLA models, boosting multi-task and spatial reasoning performance.

Chengmeng Li, Junjie Wen, Yan Peng et al.

Proposes MovieAgent, a multi-agent hierarchical CoT framework for automated movie generation, achieving superior narrative coherence and character consistency.

Weijia Wu, Zeyu Zhu, Mike Zheng Shou

COMODO enhances IMU recognition efficiency through cross-modal distillation, matching or surpassing supervised models.

Baiyu Chen, Wilson Wongso, Zechen Li et al.

ProjectEval introduces a user-interaction-based benchmark for project-level code generation evaluation, with 284 tasks and multi-level inputs, enhancing automation and explainability.

Kaiyuan Liu, Youcheng Pan, Yang Xiang et al.

HiSTF Mamba framework achieves high-fidelity text-driven motion generation with redundancy-free spatiotemporal fusion, reducing FID to 0.189 on HumanML3D.

Xingzu Zhan, Chen Xie, Honghang Chen et al.

SafePlan combines formal logic and chain-of-thought reasoning, reducing harmful prompt acceptance by 90.5% in robotic task planning.

Ike Obi, Vishnunandan L. N. Venkatesh, Weizheng Wang et al.

FEA-Bench evaluates LLMs' ability for repository-level feature development; current models resolve only ~10% of tasks.

Wei Li, Xin Zhang, Zhongxin Guo et al.

Seg-Zero uses cognitive reinforcement to generate explicit reasoning chains, achieving 57.5% zero-shot performance on ReasonSeg with a decoupled architecture.

Yuqi Liu, Bohao Peng, Zhisheng Zhong et al.

Evaluates small language models' resistance to jailbreak attacks; 61% show over 40% success rate under attack.

Zhibo Wang, Wenhui Zhang, Huiyu Xu et al.

This study proposes a taxonomy of 19 inefficiency patterns in 492 LLM-generated Python codes, validated through expert surveys.

Altaf Allah Abbassi, Leuson Da Silva, Amin Nikanjam et al.

Proposes Bayesian Fields, a task-driven open-set semantic mapping method using probabilistic 3D Gaussian representations and Bayesian updating for multi-view fusion, outperforming traditional averaging.

Dominic Maggio, Luca Carlone

LLM Agents enhance recommendation and search systems, significantly improving information retrieval efficiency.

Yu Zhang, Shutong Qiao, Jiaqi Zhang et al.

VideoPainter uses a dual-stream architecture for any-length video inpainting, enhancing semantic consistency.

Yuxuan Bian, Zhaoyang Zhang, Xuan Ju et al.

TrajectorCrafter employs dual-stream diffusion to redirect monocular video camera trajectories, integrating point cloud rendering and source video for high-fidelity results.

Mark YU, Wenbo Hu, Jinbo Xing et al.

CompDiffuser leverages bidirectional diffusion models for compositional long-horizon trajectory stitching, outperforming existing methods.

Yunhao Luo, Utkarsh A. Mishra, Yilun Du et al.