ИСТИНА

Methods and apparatuses are described for obtaining scores for sets of weights within layers of a neural network. Such a score is based on a set of weights in a first layer and one or more of (i) an (optional) normalization coefficient in a subsequent normalization layer and (ii) a set of weights within a second layer. Based on the score the set of weights may be removed from the layer. Such a pruning may reduce the size of the respective layer. Based on a group score, groups of sets of weights may be pruned from one or more layers of the neural network. The size of the groups can be based, for example, on hardware specifications.