New research proposes method to measure learned equivariance in ML models

A new research paper proposes a method for measuring learned equivariance in machine learning models, particularly for intermediate representations where group actions are not explicitly defined. The study introduces a test based on normalized mutual information (NMI) for discrete settings and utilizes centered kernel alignment (CKA) for high-dimensional activations, offering a scalable approach with guarantees for orthogonal feature transformations. The findings suggest that data augmentation generally enhances transformation consistency, allowing for direct probing of learned equivariance from representations. AI

IMPACT Provides a new framework for evaluating how well machine learning models capture and utilize symmetries in data.

RANK_REASON The cluster contains a single academic paper detailing a new methodology for machine learning research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Claire Schlesinger
• DagsHub
• Gotit.pub
• Hugging Face
• IArxiv
• Influence Flower
• machine learning
• ScienceCast

• paper
• model release

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