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。
37 篇论文 · 按点赞排序
Hao-Wen Dong, Xiaoyu Liu, Jordi Pons +5 authors
A conditional diffusion model leverages pretrained CLIP and diffusion prior models to generate audio from text queries using unlabeled videos, offering competitive performance in text-to-audio and image-to-audio synthesis.
Stephen Casper, Jason Lin, Joe Kwon +2 authors
A red teaming framework is proposed to identify and refine undesired behaviors in large language models like GPT-2 and GPT-3, using human evaluations and constructing a labeled dataset of common-knowledge statements.
Christopher T. Small, Ivan Vendrov, Esin Durmus +6 authors
Large Language Models enhance Polis conversations through summarization, but their context limitations pose risks that need to be managed.
Chun-Hsiao Yeh, Bryan Russell, Josef Sivic +2 authors
Meta-personalization of large-scale vision-language models enables instance-level video search without explicit supervision, using a personal video instance retrieval benchmark and automatic identification of visual instances.
Yi-Lun Liao, Brandon Wood, Abhishek Das +1 authors
Equivariant Transformers, specifically EquiformerV2, achieve superior performance in 3D atomistic systems by using eSCN convolutions and architectural improvements, demonstrating better scalability, speed, and accuracy.
Wei-Tsung Lu, Ju-Chiang Wang, Yun-Ning Hung
A novel deep neural network, Perceiver TF, improves multitrack music transcription by integrating a hierarchical expansion and Transformer layer into the Perceiver architecture, achieving better scalability and performance compared to existing methods.
Yuxiang Yang, Guanya Shi, Xiangyun Meng +4 authors
A novel hierarchical framework using reinforcement learning and optimal control enables a legged robot to perform continuous adaptive jumps with high versatility and robustness.
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