Artificial intelligence systems pervade every-day life. Their application to embedded scenarios comes with many limitations caused by computational capability constraints. We work to produce lightweight NN solutions while maintaining high performances. This talk is part of the annual review of the PhD research work.