Presenting PB-EDropout at ESANN 2026

From 22 to 24 April 2026, I was in Bruges for ESANN, the European Symposium on Artificial Neural Networks, to present our paper Energy-Based Dropout with Patch-Level Regularization.

In one sentence: instead of switching neurons off at random, PB-EDropout keeps the most relevant patches of each feature map and drops the rest, which improves accuracy over standard dropout and shows which regions the network relies on.

Joint work with Bilal Faye, Djamel Bouchaffra, Nadjib Lazaar, Hanane Azzag and Mustapha Lebbah. Between sessions, there was still time for a walk along the canals.