AAAI 2026Published

TIDE: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation

Victor Shea-Jay Huang^{* (equal contribution)}, Le Zhuo^{* (equal contribution)}, Yi Xin, Zhaokai Wang, Fu-Yun Wang, Yuchi Wang, Renrui Zhang, Peng Gao, Hongsheng Li

AAAI arXiv Citation

TIDE visualizes spatial, semantic, and class-level sparse features for generated house and bird images.

TL;DR

Temporal-aware sparse autoencoders reveal hierarchical concepts across diffusion timesteps, making diffusion transformers easier to interpret and control.用时间感知的稀疏自编码器提取扩散各时间步中的层次化概念，理解并控制 Diffusion Transformer 的生成过程。

Citation

Download .bib

@article{huang2026tide,
  title = {{TIDE}: Temporal-Aware Sparse Autoencoders for Interpretable Diffusion Transformers in Image Generation},
  author = {Huang, Victor Shea-Jay and Zhuo, Le and Xin, Yi and Wang, Zhaokai and Wang, Fu-Yun and Wang, Yuchi and Zhang, Renrui and Gao, Peng and Li, Hongsheng},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume = {40},
  number = {1},
  pages = {435--443},
  year = {2026},
  doi = {10.1609/aaai.v40i1.37006},
  url = {https://ojs.aaai.org/index.php/AAAI/article/view/37006}
}