Recurrent Equivariant Constraint Modulation: Learning Per-Layer Symmetry Relaxation from Data

Stefanos Pertigkiozoglou, Mircea Petrache, Shubhendu Trivedi, Kostas Daniilidis · 2026

Read Recurrent Equivariant Constraint Modulation: Learning Per-Layer Symmetry Relaxation from Data by Stefanos Pertigkiozoglou, Mircea Petrache, Shubhendu Trivedi, Kostas Daniilidis (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02853.txt.

Description

Stefanos Pertigkiozoglou and 3 co-authors study how neural networks should handle symmetry in a 2026 paper. Networks that respect a task's symmetry often generalize better. Forcing perfect symmetry, though, can make training harder when the data is only nearly symmetric.

Their method, RECM, lets each layer learn how much to relax its symmetry rule. The learning uses only the training signal and each layer's own symmetry properties. The authors prove the relaxation settles near a measured symmetry gap. Layers with exact symmetry return to full symmetry. Layers with rough symmetry keep useful flexibility. Tests on exact and approximate symmetry tasks, including molecule generation, favor RECM.

Catalog record

CREATORStefanos Pertigkiozoglou, Mircea Petrache, Shubhendu Trivedi, Kostas Daniilidis
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE59 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02853

Browse the full library · About the Tower of Babel