ICML
2005
IEEE
15 years 1 months ago
2005
IEEE
We extend the PAC-Bayes theorem to the sample-compression setting where each classifier is represented by two independent sources of information: a compression set which consists ...