Interpreting CLIP with Hierarchical Sparse Autoencoders

Vladimir Zaigrajew, Hubert Baniecki, Przemyslaw Biecek
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:73918-73956, 2025.

Abstract

Sparse autoencoders (SAEs) are useful for detecting and steering interpretable features in neural networks, with particular potential for understanding complex multimodal representations. Given their ability to uncover interpretable features, SAEs are particularly valuable for analyzing vision-language models (e.g., CLIP and SigLIP), which are fundamental building blocks in modern large-scale systems yet remain challenging to interpret and control. However, current SAE methods are limited by optimizing both reconstruction quality and sparsity simultaneously, as they rely on either activation suppression or rigid sparsity constraints. To this end, we introduce Matryoshka SAE (MSAE), a new architecture that learns hierarchical representations at multiple granularities simultaneously, enabling a direct optimization of both metrics without compromise. MSAE establishes a state-of-the-art Pareto frontier between reconstruction quality and sparsity for CLIP, achieving 0.99 cosine similarity and less than 0.1 fraction of variance unexplained while maintaining 80% sparsity. Finally, we demonstrate the utility of MSAE as a tool for interpreting and controlling CLIP by extracting over 120 semantic concepts from its representation to perform concept-based similarity search and bias analysis in downstream tasks like CelebA. We make the codebase available at https://github.com/WolodjaZ/MSAE.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-zaigrajew25a, title = {Interpreting {CLIP} with Hierarchical Sparse Autoencoders}, author = {Zaigrajew, Vladimir and Baniecki, Hubert and Biecek, Przemyslaw}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {73918--73956}, year = {2025}, editor = {Singh, Aarti and Fazel, Maryam and Hsu, Daniel and Lacoste-Julien, Simon and Berkenkamp, Felix and Maharaj, Tegan and Wagstaff, Kiri and Zhu, Jerry}, volume = {267}, series = {Proceedings of Machine Learning Research}, month = {13--19 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v267/main/assets/zaigrajew25a/zaigrajew25a.pdf}, url = {https://proceedings.mlr.press/v267/zaigrajew25a.html}, abstract = {Sparse autoencoders (SAEs) are useful for detecting and steering interpretable features in neural networks, with particular potential for understanding complex multimodal representations. Given their ability to uncover interpretable features, SAEs are particularly valuable for analyzing vision-language models (e.g., CLIP and SigLIP), which are fundamental building blocks in modern large-scale systems yet remain challenging to interpret and control. However, current SAE methods are limited by optimizing both reconstruction quality and sparsity simultaneously, as they rely on either activation suppression or rigid sparsity constraints. To this end, we introduce Matryoshka SAE (MSAE), a new architecture that learns hierarchical representations at multiple granularities simultaneously, enabling a direct optimization of both metrics without compromise. MSAE establishes a state-of-the-art Pareto frontier between reconstruction quality and sparsity for CLIP, achieving 0.99 cosine similarity and less than 0.1 fraction of variance unexplained while maintaining 80% sparsity. Finally, we demonstrate the utility of MSAE as a tool for interpreting and controlling CLIP by extracting over 120 semantic concepts from its representation to perform concept-based similarity search and bias analysis in downstream tasks like CelebA. We make the codebase available at https://github.com/WolodjaZ/MSAE.} }
Endnote
%0 Conference Paper %T Interpreting CLIP with Hierarchical Sparse Autoencoders %A Vladimir Zaigrajew %A Hubert Baniecki %A Przemyslaw Biecek %B Proceedings of the 42nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2025 %E Aarti Singh %E Maryam Fazel %E Daniel Hsu %E Simon Lacoste-Julien %E Felix Berkenkamp %E Tegan Maharaj %E Kiri Wagstaff %E Jerry Zhu %F pmlr-v267-zaigrajew25a %I PMLR %P 73918--73956 %U https://proceedings.mlr.press/v267/zaigrajew25a.html %V 267 %X Sparse autoencoders (SAEs) are useful for detecting and steering interpretable features in neural networks, with particular potential for understanding complex multimodal representations. Given their ability to uncover interpretable features, SAEs are particularly valuable for analyzing vision-language models (e.g., CLIP and SigLIP), which are fundamental building blocks in modern large-scale systems yet remain challenging to interpret and control. However, current SAE methods are limited by optimizing both reconstruction quality and sparsity simultaneously, as they rely on either activation suppression or rigid sparsity constraints. To this end, we introduce Matryoshka SAE (MSAE), a new architecture that learns hierarchical representations at multiple granularities simultaneously, enabling a direct optimization of both metrics without compromise. MSAE establishes a state-of-the-art Pareto frontier between reconstruction quality and sparsity for CLIP, achieving 0.99 cosine similarity and less than 0.1 fraction of variance unexplained while maintaining 80% sparsity. Finally, we demonstrate the utility of MSAE as a tool for interpreting and controlling CLIP by extracting over 120 semantic concepts from its representation to perform concept-based similarity search and bias analysis in downstream tasks like CelebA. We make the codebase available at https://github.com/WolodjaZ/MSAE.
APA
Zaigrajew, V., Baniecki, H. & Biecek, P.. (2025). Interpreting CLIP with Hierarchical Sparse Autoencoders. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:73918-73956 Available from https://proceedings.mlr.press/v267/zaigrajew25a.html.

Related Material