arXiv preprint / cs.CV

Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes

作者：Junlin Han, Shengbang Tong, David Fan, Minghao Chen, Philip Torr, Filippos Kokkinos, Mike Lewis · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-08-06

2026-08-06视觉语言生成模型强化学习
新标签打开 AlphaXiv新标签打开 Project
Towards Physics of Multimodal Pretraining: Knowledge Flow, Modality Synergy, Early Unification, and Recipes figure
AlphaXiv 中文概览（可滚动查看）