bibtype C - Conference Paper (international conference)
ARLID 0645094
utime 20260126101547.5
mtime 20260126235959.9
title (primary) (eng) Harmformer: Harmonic Networks Meet Transformers for Continuous Roto-Translation Equivariance
specification
page_count 27 s.
media_type E
serial
ARLID cav_un_epca*0645102
ISSN 2640-3498
title NeurIPS 2024 Workshop on Symmetry and Geometry in Neural Representations
publisher
place San Diego
name ML Research Press
year 2026
keyword transformers
keyword roto-translation
keyword geometric deep learning
keyword invariance
keyword equivariance
keyword robustness
author (primary)
ARLID cav_un_auth*0438860
name1 Karella
name2 Tomáš
institution UTIA-B
full_dept (cz) Zpracování obrazové informace
full_dept (eng) Department of Image Processing
department (cz) ZOI
department (eng) ZOI
country CZ
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
author
ARLID cav_un_auth*0439197
name1 Harmanec
name2 Adam
institution UTIA-B
full_dept (cz) Zpracování obrazové informace
full_dept Department of Image Processing
department (cz) ZOI
department ZOI
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
author
ARLID cav_un_auth*0293863
name1 Kotera
name2 Jan
institution UTIA-B
full_dept (cz) Zpracování obrazové informace
full_dept Department of Image Processing
department (cz) ZOI
department ZOI
full_dept Department of Image Processing
country CZ
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
author
ARLID cav_un_auth*0254045
name1 Blažek
name2 Jan
institution UTIA-B
full_dept (cz) Zpracování obrazové informace
full_dept Department of Image Processing
department (cz) ZOI
department ZOI
full_dept Department of Image Processing
country CZ
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
author
ARLID cav_un_auth*0101209
name1 Šroubek
name2 Filip
institution UTIA-B
full_dept (cz) Zpracování obrazové informace
full_dept Department of Image Processing
department (cz) ZOI
department ZOI
full_dept Department of Image Processing
fullinstit Ústav teorie informace a automatizace AV ČR, v. v. i.
source
source_type pdf
source_size 11MB
url https://library.utia.cas.cz/separaty/2026/ZOI/sroubek-0645094.pdf
cas_special
project
project_id GA24-10069S
agency GA ČR
country CZ
ARLID cav_un_auth*0472834
project
project_id VJ02010029
agency GA MV
country CZ
ARLID cav_un_auth*0449227
abstract (eng) Convolutional Neural Networks exhibit inherent equivariance to image translation, leading to efficient parameter and data usage, faster learning, and improved robustness. The concept of translation equivariant networks has been successfully extended to rotation transformation using group convolution for discrete rotation groups and harmonic functions for the continuous rotation group encompassing 360°. We explore the compatibility of the Self-Attention mechanism with full rotation equivariance, in contrast to previous studies that focused on discrete rotation. We introduce the Harmformer, a harmonic transformer with a convolutional stem that achieves equivariance for both translation and continuous rotation. Accompanied by an end-to-end equivariance proof, the Harmformer not only outperforms previous equivariant transformers, but also demonstrates inherent stability under any continuous rotation, even without seeing rotated samples during training.
action
ARLID cav_un_auth*0502239
name NeurIPS 2024 Workshop on Symmetry and Geometry in Neural Representations
dates 20241214
mrcbC20-s 20241215
place Vancouver
country CA
RIV JD
reportyear 2026
presentation_type PR
inst_support RVO:67985556
permalink https://hdl.handle.net/11104/0374942
confidential S
mrcbC96 https://arxiv.org/pdf/2411.03794
arlyear 2026
mrcbU14 SCOPUS
mrcbU24 PUBMED
mrcbU34 WOS
mrcbU63 cav_un_epca*0645102 NeurIPS 2024 Workshop on Symmetry and Geometry in Neural Representations ML Research Press 2026 San Diego 2640-3498 2640-3498