AI RESEARCH
Soft Equivariance Regularization for Invariant Self-Supervised Learning
arXiv CS.LG
•
ArXi:2603.06693v1 Announce Type: cross Self-supervised learning (SSL) typically learns representations invariant to semantic-preserving augmentations. While effective for recognition, enforcing strong invariance can suppress transformation-dependent structure that is useful for robustness to geometric perturbations and spatially sensitive transfer. A growing body of work, therefore, augments invariance-based SSL with equivariance objectives, but these objectives are often imposed on the same final representation.