With a strong enough non-equivariance penalty, regularized augmented gradient flow converges exponentially fast to the equivariant subspace.
Steerable cnns,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Data Augmentation and Regularization for Learning Group Equivariance
With a strong enough non-equivariance penalty, regularized augmented gradient flow converges exponentially fast to the equivariant subspace.