PuSH - Publikationsserver des Helmholtz Zentrums München

Lim, H. ; Choi, J.* ; Choo, J.* ; Schneider, S.

Sparse autoencoders reveal selective remapping of visual concepts during adaptation.

In: (13th International Conference on Learning Representations Iclr 2025, 24 - 28 April 2025, Singapur). 2025. 46012-46037 (13th International Conference on Learning Representations Iclr 2025)
Postprint
Adapting foundation models for specific purposes has become a standard approach to build machine learning systems for downstream applications. Yet, it is an open question which mechanisms take place during adaptation. Here we develop a new Sparse Autoencoder (SAE) for the CLIP vision transformer, named PatchSAE, to extract interpretable concepts at granular levels (e.g., shape, color, or semantics of an object) and their patch-wise spatial attributions. We explore how these concepts influence the model output in downstream image classification tasks and investigate how recent state-of-the-art prompt-based adaptation techniques change the association of model inputs to these concepts. While activations of concepts slightly change between adapted and non-adapted models, we find that the majority of gains on common adaptation tasks can be explained with the existing concepts already present in the non-adapted foundation model. This work provides a concrete framework to train and use SAEs for Vision Transformers and provides insights into explaining adaptation mechanisms.
Weitere Metriken?
[➜Einloggen]
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Korrespondenzautor
ISSN (print) / ISBN [9798331320850]
Konferenztitel 13th International Conference on Learning Representations Iclr 2025
Konferzenzdatum 24 - 28 April 2025
Konferenzort Singapur
Quellenangaben Band: , Heft: , Seiten: 46012-46037 Artikelnummer: , Supplement: ,
Nichtpatentliteratur Publikationen
Institut(e) Institute of Computational Biology (ICB)