S-GAI initializes sigmoidal MLPs by turning class-specific SVD principal directions into pairs of sigmoid gates, yielding stronger initial performance than Xavier on MNIST, Fashion-MNIST, and CIFAR-10.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
S-GAI: Spectral Geometry-Aware Initialization for Sigmoidal MLPs -- From Dataset Geometry to Network Weights
S-GAI initializes sigmoidal MLPs by turning class-specific SVD principal directions into pairs of sigmoid gates, yielding stronger initial performance than Xavier on MNIST, Fashion-MNIST, and CIFAR-10.