Training neural networks with PAC-Bayes objectives yields MNIST test error of 1.4% and a non-vacuous risk bound of 2.3%, much tighter than prior PAC-Bayes certificates.
Self bounding learning algorithms
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
PAC-Bayes with Backprop
Training neural networks with PAC-Bayes objectives yields MNIST test error of 1.4% and a non-vacuous risk bound of 2.3%, much tighter than prior PAC-Bayes certificates.