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。
50 篇论文 · 按点赞排序
Jiabin Hua, Hengyuan Xu, Aojie Li +4 authors
A diffusion framework called PixelSmile is introduced that disentangles facial expression semantics through symmetric joint training and contrastive learning to enable precise, controllable, and fine-grained expression editing with robust identity preservation.
Songchun Zhang, Zeyue Xue, Siming Fu +6 authors
Astrolabe is an efficient online reinforcement learning framework for distilled autoregressive video models that improves generation quality through forward-process RL formulation and streaming training with multi-reward objectives.
Shengbang Tong, David Fan, John Nguyen +18 authors
Controlled multimodal pretraining experiments reveal key insights about unified visual representations, data complementarity, world modeling emergence, and efficient scaling through mixture-of-experts architectures.
Zhuofeng Li, Dongfu Jiang, Xueguang Ma +7 authors
OpenResearcher presents a reproducible pipeline for training deep research agents using offline search environments and synthesized trajectories, achieving improved accuracy on benchmark tasks.
Weinan Dai, Hanlin Wu, Qiying Yu +13 authors
CUDA Agent, a large-scale agentic reinforcement learning system, achieves state-of-the-art performance in CUDA kernel optimization by combining scalable data synthesis, skill-augmented development environment, and reinforcement learning techniques.
Xiangru Jian, Shravan Nayak, Kevin Qinghong Lin +5 authors
CUA-Suite introduces a large-scale ecosystem of expert video demonstrations and annotations for computer-use agents, providing continuous screen recordings and detailed reasoning annotations to advance desktop automation capabilities.
Xianjin Wu, Dingkang Liang, Tianrui Feng +5 authors
A video diffusion model is repurposed as a latent world simulator to enhance multimodal large language models with implicit 3D structural priors and physical laws through spatiotemporal feature extraction and semantic integration.
Zhongxing Xu, Zhonghua Wang, Zhe Qian +10 authors
Latent Entropy-Aware Decoding (LEAD) enhances multimodal large reasoning models by using entropy-aware reasoning mode switching and prior-guided visual anchor injection to reduce hallucinations.
Yuan Liang, Ruobin Zhong, Haoming Xu +46 authors
SkillNet introduces an open infrastructure for systematically accumulating and transferring AI skills through a unified ontology, significantly improving agent performance across multiple domains.
Hongtao Wu, Boyun Zheng, Dingjie Song +5 authors
Medical AI Scientist represents the first autonomous research framework designed for clinical applications, enabling evidence-based hypothesis generation and manuscript drafting through clinician-engineer collaboration across three research modes.
Zhen Li, Zian Meng, Shuwei Shi +5 authors
WildWorld is a large-scale dataset for action-conditioned world modeling that provides explicit state annotations from a photorealistic game, enabling better understanding of latent-state dynamics and long-horizon consistency.
Junjie Li, Xinrui Guo, Yuhao Wu +3 authors
Large language models struggle with maintaining narrative consistency over long-form storytelling, exhibiting predictable patterns of contradictions that this benchmark identifies and categorizes.
Ibragim Badertdinov, Maksim Nekrashevich, Anton Shevtsov +1 authors
A large-scale dataset of software engineering tasks spanning multiple programming languages and repositories was created using an automated pipeline that generates executable environments and filters unreliable instances through LLM validation.
Nimrod Shabtay, Moshe Kimhi, Artem Spector +5 authors
AwaRes is a spatial-on-demand framework for vision-language models that dynamically retrieves high-resolution image segments based on query needs, using tool-calling and multi-turn reinforcement learning with composite rewards.
Fangfu Liu, Diankun Wu, Jiawei Chi +7 authors
Spatial-TTT enables streaming visual-based spatial intelligence through test-time training that adapts parameters to capture spatial evidence over long video sequences using hybrid architecture and 3D spatiotemporal convolution.
Zonglin Yang, Lidong Bing
MOOSE-Star framework enables efficient training and inference for generative reasoning by addressing combinatorial complexity through decomposed subtasks, hierarchical search, and bounded composition.
Wei Yu, Runjia Qian, Yumeng Li +8 authors
Video diffusion models use hybrid spatial memory to maintain consistency under camera motion and enable long-term scene editing and navigation.
Yuanyuan Gao, Hao Li, Yifei Liu +14 authors
Holi-Spatial presents the first fully automated, large-scale, spatially-aware multimodal dataset constructed from raw video inputs, supporting multi-level spatial supervision for 3D scene understanding and spatial reasoning tasks.
Zimo Wen, Boxiu Li, Wanbo Zhang +11 authors
Unified multimodal models show mixed performance in generation-to-understanding tasks, with specific subtasks benefiting from enhanced spatial and reasoning capabilities while overall performance lags behind specialized vision-language models.
Lianghui Zhu, Yuxin Fang, Bencheng Liao +10 authors
Scaling depth is a key driver for large language models (LLMs). Yet, as LLMs become deeper, they often suffer from signal degradation: informative features formed in shallow layers are gradually diluted by repeated residual updates, making them harder to recover in deeper layers. We introduce mixture-of-depths attention (MoDA), a mechanism that allows each attention head to attend to sequence KV pairs at the current layer and depth KV pairs from preceding layers. We further describe a hardware-efficient algorithm for MoDA that resolves non-contiguous memory-access patterns, achieving 97.3% of FlashAttention-2's efficiency at a sequence length of 64K. Experiments on 1.5B-parameter models demonstrate that MoDA consistently outperforms strong baselines. Notably, it improves average perplexity by 0.2 across 10 validation benchmarks and increases average performance by 2.11% on 10 downstream tasks, with a negligible 3.7% FLOPs computational overhead. We also find that combining MoDA with post-norm yields better performance than using it with pre-norm. These results suggest that MoDA is a promising primitive for depth scaling. Code is released at https://github.com/hustvl/MoDA .
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