Training big artificial neural nets in order to apply them for solving problems can take a lot of computation resources, which need a lot of energy with a big carbon footprint. By pruning the connectivity in an intelligent way after initial steps, the energy consumption can be cut to a tenth.