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。
41 篇论文 · 按点赞排序
Haoran Cheng, Liang Peng, Linxuan Xia +5 authors
This method enhances text-to-video synthesis by using a search-based generation pipeline with motion priors from existing datasets, improving video realism without extensive retraining.
Rafael Rafailov, Yaswanth Chittepu, Ryan Park +5 authors
The study investigates reward over-optimization and hacking in Direct Alignment Algorithms, finding similarities to classical Reinforcement Learning from Human Feedback methods, and formalizes this problem for DAAs.
Benjamin Thérien, Charles-Étienne Joseph, Boris Knyazev +3 authors
Maximal Update Parametrization improves the meta-generalization of learned optimizers, allowing zero-shot generalization to larger and deeper models with reduced computational cost.
Li Yang, Anushya Subbiah, Hardik Patel +4 authors
The Item-Language Model integrates user interaction signals into language-aligned item representations to enhance recommendations while preserving pre-trained language abilities.
Soroush Nasiriany, Abhiram Maddukuri, Lance Zhang +5 authors
RoboCasa is a large-scale simulation framework using realistic physical environments, diverse tasks, and datasets enhanced with generative AI for training generalist robots.
Shaoshu Yang, Yong Zhang, Xiaodong Cun +2 authors
A training-free video interpolation method enhances frame rate in generative diffusion models by incorporating hidden state correction modules within a self-cascaded architecture, achieving performance comparable to models trained with extensive computational resources.
Cong Wang, Kuan Tian, Jun Zhang +7 authors
V-Express addresses the challenge of weak control signals in portrait video generation through progressive training and conditional dropout, enabling effective audio-controlled video generation alongside other conditions.
Sike Wang, Jia Li, Pan Zhou +1 authors
4-bit quantized Shampoo, a second-order optimizer, achieves comparable performance to 32-bit Shampoo while using less memory, by optimizing the eigenvector matrix quantization and enhancing orthogonality.
Rongzhi Zhang, Jiaming Shen, Tianqi Liu +7 authors
PLaD is a preference-based distillation framework that re-calibrates compact student LLMs by comparing them to larger teacher models without direct access to internal states.
Yichuan Wang, Yang Liu, Yu Yan +2 authors
A compact Xmodel-LM language model, trained on a 2 trillion-token dataset, outperforms similar-sized open-source models.
Dejia Xu, Weili Nie, Chao Liu +4 authors
CamCo enhances camera pose control in video generation by using Plücker coordinates and an epipolar attention module, improving 3D consistency and object motion synthesis.
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