Steerable Transformers for Volumetric Data

Soumyabrata Kundu, Risi Kondor
Proceedings of the 42nd International Conference on Machine Learning, PMLR 267:32011-32031, 2025.

Abstract

We introduce Steerable Transformers, an extension of the Vision Transformer mechanism that maintains equivariance to the special Euclidean group $\mathrm{SE}(d)$. We propose an equivariant attention mechanism that operates on features extracted by steerable convolutions. Operating in Fourier space, our network utilizes Fourier space non-linearities. Our experiments in both two and three dimensions show that adding steerable transformer layers to steerable convolutional networks enhances performance.

Cite this Paper

BibTeX
@InProceedings{pmlr-v267-kundu25a, title = {Steerable Transformers for Volumetric Data}, author = {Kundu, Soumyabrata and Kondor, Risi}, booktitle = {Proceedings of the 42nd International Conference on Machine Learning}, pages = {32011--32031}, 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/kundu25a/kundu25a.pdf}, url = {https://proceedings.mlr.press/v267/kundu25a.html}, abstract = {We introduce Steerable Transformers, an extension of the Vision Transformer mechanism that maintains equivariance to the special Euclidean group $\mathrm{SE}(d)$. We propose an equivariant attention mechanism that operates on features extracted by steerable convolutions. Operating in Fourier space, our network utilizes Fourier space non-linearities. Our experiments in both two and three dimensions show that adding steerable transformer layers to steerable convolutional networks enhances performance.} }
Endnote
%0 Conference Paper %T Steerable Transformers for Volumetric Data %A Soumyabrata Kundu %A Risi Kondor %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-kundu25a %I PMLR %P 32011--32031 %U https://proceedings.mlr.press/v267/kundu25a.html %V 267 %X We introduce Steerable Transformers, an extension of the Vision Transformer mechanism that maintains equivariance to the special Euclidean group $\mathrm{SE}(d)$. We propose an equivariant attention mechanism that operates on features extracted by steerable convolutions. Operating in Fourier space, our network utilizes Fourier space non-linearities. Our experiments in both two and three dimensions show that adding steerable transformer layers to steerable convolutional networks enhances performance.
APA
Kundu, S. & Kondor, R.. (2025). Steerable Transformers for Volumetric Data. Proceedings of the 42nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 267:32011-32031 Available from https://proceedings.mlr.press/v267/kundu25a.html.

Related Material