arXiv preprint / cs.CV

D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders

作者：Xinyue Xu, Jiahao Zhang, Lijie Hu, Peter Hase, Hao Wang · 单位：Pivotal Research Mohamed bin Zayed University of Artificial Intelligence, Schmidt Sciences Stanford University University of Illinois at Urbana-Champaign · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-01

2026-10-01生成模型
新标签打开 AlphaXiv新标签打开 Project
D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders figure
AlphaXiv 中文概览（可滚动查看）