Rotation-augmented CNNs learn equivariance easily for denoising and prediction, but only large models keep generation outputs invariant to seed rotations, and their latent codes do not identify rotated molecules as the same object.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Do we need equivariant models for molecule generation?
Rotation-augmented CNNs learn equivariance easily for denoising and prediction, but only large models keep generation outputs invariant to seed rotations, and their latent codes do not identify rotated molecules as the same object.