Some theoretical improvements on the tightness of PAC-Bayes risk certificates for neural networks

Open in new window