Abstract: Filter pruning has been widely applied to neural network compression and acceleration. Existing methods usually utilize pre-defined pruning criteria, such as Lp-norm, to prune unimportant ...