PAC-Bayes Analysis of Multi-view Learning

This paper presents four PAC-Bayes bounds to analyse the generalisation performance of multi-view classifiers. These bounds adopt data dependent Gaussian priors which emphasize the classifiers with high view agreements. The centre of the prior for the first two bounds is the origin, while the centre of the prior for the last two bounds is given by a data dependent vector. Another important ingredient to obtain these bounds is two derived logarithmic determinant inequalities whose difference lies at whether the dimensionality of data is involved. We evaluate the multi-view PAC-Bayes bounds on benchmark data with preliminary experimental results indicating their usefulness.