New framework tackles symmetry increase in Equivariant Neural Networks

Researchers have developed a new framework to address the issue of symmetry increase in Equivariant Neural Networks (ENNs). This phenomenon occurs when ENNs, despite their ability to represent geometric structures, exhibit increased output symmetry beyond the input's original symmetries, thus degrading their expressivity. The paper provides a theoretical characterization of this symmetry increase, proving that it has a computable infimum based on the feature space structure. Practical guidelines and an algorithm are proposed to effectively reduce this increase, with experimental validation on synthetic data and the QM9 dataset. AI

IMPACT Improves the expressivity and applicability of Equivariant Neural Networks in scientific domains.

RANK_REASON Academic paper detailing a new theoretical framework and algorithm for reducing symmetry increase in Equivariant Neural Networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• Influence Flower
• QM9
• ScienceCast

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →