Explore · 每周精选
发现最受关注的研究论文,追踪研究趋势,订阅感兴趣的期刊与关键词。
Maijunxian Wang, Ruisi Wang, Juyi Lin +53 authors
A large-scale video reasoning dataset and benchmark are introduced to study video intelligence capabilities beyond visual quality, enabling systematic analysis of spatiotemporal reasoning and generalization across diverse tasks.
50 篇论文 · 按点赞排序
Zixuan Huang, Xin Xia, Yuxi Ren +11 authors
Large reasoning models can implicitly determine optimal stopping points for thinking, which SAGE-RL enhances by incorporating efficient reasoning patterns into pass@1 inference for improved accuracy and efficiency.
Guobin Shen, Chenxiao Zhao, Xiang Cheng +2 authors
VESPO addresses training instability in LLM reinforcement learning by using variational formulation with variance reduction to correct policy divergence without length normalization.
Jingxuan Wei, Siyuan Li, Yuhang Xu +21 authors
World Models require three consistency principles—modal, spatial, and temporal—for general artificial intelligence, with a proposed benchmark evaluating multimodal learning systems.
Hongrui Jia, Chaoya Jiang, Shikun Zhang +1 authors
Diagnostic-driven Progressive Evolution enables continuous improvement of large multimodal models through iterative diagnosis and targeted data generation guided by identified weaknesses.
Zhiheng Song, Jingshuai Zhang, Chuan Qin +6 authors
MobileBench is a scalable benchmark for evaluating LLM-based route-planning agents in real-world scenarios, featuring anonymized user queries and a deterministic sandbox for reproducible testing.
Renjie Pi, Grace Lam, Mohammad Shoeybi +3 authors
Researchers developed a synthetic task generation pipeline and analyzed data strategies to improve terminal agent performance, creating a large-scale dataset and models that outperform larger counterparts on benchmark tests.
Yuqing Li, Jiangnan Li, Mo Yu +4 authors
A lightweight reranking framework uses attention scores from selected heads to estimate passage-query relevance, achieving strong performance across multiple domains and benchmarks.
Jiayu Wang, Yifei Ming, Zixuan Ke +3 authors
SkillOrchestra presents a skill-aware orchestration framework that improves compound AI system performance through fine-grained skill modeling and efficient agent selection, achieving superior results with significantly reduced learning costs compared to reinforcement learning-based methods.
Yongtong Wu, Shaoyuan Chen, Yinmin Zhong +10 authors
DualPath addresses KV-cache storage I/O bottlenecks in multi-turn LLM inference by introducing dual-path loading and dynamic load balancing across prefill and decode engines.
Lei Xin, Yuhao Zheng, Ke Cheng +3 authors
HyTRec addresses the challenge of modeling long user behavior sequences by combining linear and softmax attention mechanisms with a temporal-aware delta network to balance efficiency and retrieval precision.
Guibin Chen, Dixuan Lin, Jiangping Yang +46 authors
SkyReels V4 is a unified multimodal video foundation model that generates, edits, and inpaints video and audio simultaneously using a dual-stream architecture with shared text encoding and efficient high-resolution processing.
Hojung Jung, Rodrigo Hormazabal, Jaehyeong Jo +5 authors
MolHIT presents a hierarchical discrete diffusion model for molecular graph generation that achieves superior chemical validity and property-guided synthesis compared to existing 1D and graph-based approaches.
Xiaoxi Li, Wenxiang Jiao, Jiarui Jin +8 authors
OmniGAIA benchmark evaluates multi-modal agents on complex reasoning tasks across video, audio, and image modalities, while OmniAtlas agent improves tool-use capabilities through hindsight-guided tree exploration and OmniDPO fine-tuning.
Xiao-Ming Wu, Bin Fan, Kang Liao +6 authors
Vision-Language-Action models are systematically analyzed and optimized through a unified framework, resulting in the VLANeXt model that achieves superior performance on benchmark tasks and demonstrates strong real-world generalization.
You Li, Chi Chen, Yanghao Li +4 authors
Research reveals that latent visual reasoning in multimodal models suffers from input-latent and latent-answer disconnects, leading to the proposal of CapImagine, a text-based approach that outperforms complex latent-space methods.
Xu Guo, Fulong Ye, Qichao Sun +7 authors
DreamID-Omni is a unified framework for controllable human-centric audio-video generation that uses a symmetric conditional diffusion transformer with dual-level disentanglement and multi-task progressive training to achieve state-of-the-art performance.
Zeyuan Liu, Jeonghye Kim, Xufang Luo +2 authors
EMPO² is a hybrid reinforcement learning framework that enhances exploration for large language model agents by integrating memory mechanisms with on- and off-policy updates, demonstrating improved performance and adaptability in complex environments.
Natalie Shapira, Chris Wendler, Avery Yen +35 authors
Autonomous language-model-powered agents in a live laboratory environment exhibited numerous security and governance vulnerabilities including unauthorized actions, information disclosure, and system takeovers during a two-week study with twenty researchers.
Junchen Liu, Sven Elflein, Or Litany +2 authors
Test-time training is reinterpreted as learned linear attention rather than memorization, offering architectural simplifications and improved efficiency.
Shitian Zhao, Shaoheng Lin, Ming Li +4 authors
PyVision-RL framework addresses interaction collapse in multimodal models through enhanced reinforcement learning techniques and efficient video processing strategies.
Georgy Savva, Oscar Michel, Daohan Lu +6 authors
Solaris is a multiplayer video world model that simulates consistent multi-view observations through a novel data collection system and staged training approach.
Kun Yang, Yuxuan Zhu, Yazhe Chen +7 authors
ManCAR is a recommendation framework that constrains latent reasoning within a collaborative manifold to prevent implausible trajectories and improve accuracy.
Linxi Xie, Lisong C. Sun, Ashley Neall +3 authors
A human-centric video world model conditioned on tracked head and hand poses is introduced, enabling dexterous interactions through a bidirectional video diffusion model trained for egocentric virtual environment generation.
Yutong Wang, Siyuan Xiong, Xuebo Liu +4 authors
AgentDropoutV2 is a test-time framework that dynamically optimizes multi-agent system information flow through error correction and pruning mechanisms without requiring retraining.
Xiaoxuan Wang, Han Zhang, Haixin Wang +11 authors
ARLArena framework analyzes training stability in agentic reinforcement learning and proposes SAMPO method for stable policy optimization across diverse tasks.
Yuhao Wu, Maojia Song, Yihuai Lan +8 authors
Current vision-language models lack capability to understand physical structures and causal constraints needed for complex, interactive 3D tasks, as demonstrated by the CHAIN benchmark evaluating structured action planning under physical constraints.
Shirui Chen, Cole Harrison, Ying-Chun Lee +6 authors
TOPReward is a probabilistically grounded temporal value function that uses pretrained video Vision-Language Models to estimate robotic task progress through internal token logits, achieving superior performance in zero-shot evaluations across diverse real-world tasks.
Sahal Shaji Mullappilly, Mohammed Irfan Kurpath, Omair Mohamed +5 authors
MediX-R1 presents an open-ended reinforcement learning framework for medical multimodal large language models that uses diverse reward signals and LLM-based evaluation to improve clinical reasoning beyond multiple-choice formats.
Qianben Chen, Tianrui Qin, King Zhu +21 authors
A deep learning framework called SMTL improves efficient long-horizon agentic search by replacing sequential reasoning with parallel evidence acquisition, achieving state-of-the-art performance across multiple research benchmarks while reducing reasoning steps by 70.7%.
北京市昌平区探索星信息技术及软件开发工作室
京ICP备2026059466号