Metraipolis › archive › dc_ee895f588695
D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders — arXiv:2609.39625
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.39625
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T12:45:37.283Z
- SHA-256 of the archived text
71c2add38f7a67faebd19904c1b107af5252cf16399877adeebe62f34371b06b- Stable URL of this page
https://metraipolis.com/documento/dc_ee895f588695
How to cite
D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders — arXiv:2609.39625 — arXiv (en), https://arxiv.org/abs/2609.39625 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ee895f588695 (Metraipolis archive dc_ee895f588695, sha256 71c2add38f7a67faebd19904c1b107af5252cf16399877adeebe62f34371b06b)
BibTeX
@misc{dcee895f588695,
title = {D-Scope: Decomposing and Steering Diffusion Transformers with Sparse Autoencoders — arXiv:2609.39625},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.39625},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ee895f588695; sha256 71c2add38f7a67faebd19904c1b107af5252cf16399877adeebe62f34371b06b},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.39625. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Sparse autoencoders (SAEs) reveal visual structure in diffusion transformers (DiTs), but interpreting a feature does not establish whether it can be used to control generation. We introduce D-Scope (Diffusion Scope), a framework that connects feature interpretation to generation control through shared visual evidence. D-Scope aggregates SigLIP~2 embeddings of highly activating image patches into visual centroids. Matching target text descriptions against these visual centroids in the shared image-text embedding space then enables retrieval of individual features without per-feature text annotations. The underlying patches provide evidence for inspecting each selection, while spatially masked interventions test the corresponding decoder direction at varying strengths under fixed generation conditions. We characterize 150 SAEs across two model families and five layers, and introduce a benchmark of 100 target concepts with ten contexts each spanning under-specified and explicit-conflict conditions. Our empirical results show that high reconstruction fidelity can coexist with low dictionary utilization and limited visual-evidence coverage. Under per-case best-of-sweep strength selection, contrastive retrieval yields larger mean regional SigLIP~2 gains than direct retrieval across the tested steering configurations, without consistently improving outside-region preservation. D-Scope provides an inspectable framework for evaluating sparse DiT features through their visual evidence and the effects of their decoder directions on generation. The demo is available at this https URL . Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-01 in cs.AI). Categorie arXiv: Computer Vision and Pattern Recognition (cs.CV) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.39625 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T12:14:14.122Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).