Generalization gap prediction on MNIST

0.551Kendall Correlation

Fourier generalization measure

Updated 1mo ago

Evaluation Results

Method	Links
Fourier generalization measure 2026.06		0.551
PAC-Bayes flatness 2026.06		0.546
PAC-Bayes init 2026.06		0.516
PAC-Bayes origin 2026.06		0.516
inverse margin 2026.06		0.386
PAC-Bayes magnitude flatness 2026.06		0.364
PAC-Bayes magnitude init 2026.06		0.111
PAC-Bayes magnitude origin 2026.06		0.046
path norm margin 2026.06		-0.02
path norm 2026.06		-0.046
distance spectral init 2026.06		-0.085
Weight watcher 2026.06		-0.218
log-prod Frobenius margin 2026.06		-0.242
log-sum Frobenius margin 2026.06		-0.242
log-prod Frobenius 2026.06		-0.268
log-sum Frobenius 2026.06		-0.268
L2-distance 2026.06		-0.294
Frobenius distance 2026.06		-0.294
L2-norm 2026.06		-0.307
parameter norm 2026.06		-0.307
# of parameters 2026.06		-0.327
log-prod spectral 2026.06		-0.333
log-prod spectral margin 2026.06		-0.333
log spectral init 2026.06		-0.333
Frobenius spectral 2026.06		-0.333
log spectral origin 2026.06		-0.333
log-sum spectral margin 2026.06		-0.333
log-sum spectral 2026.06		-0.333