Using DFT magnitude as an autoencoder reconstruction target improves CIFAR-10 linear-probe accuracy over raw-pixel reconstruction, from 35.29% to 46.05% validation top-1.
Representation Learning: A Review and New Perspectives
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.NE 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Employing Discrete Fourier Transform in Representational Learning
Using DFT magnitude as an autoencoder reconstruction target improves CIFAR-10 linear-probe accuracy over raw-pixel reconstruction, from 35.29% to 46.05% validation top-1.