Research into impact of channel comparison method on efficiency of algorithms for channel-wise pruning of convolutional neural networks

The paper is devoted to solving the problem of pruning a neural network, the purpose of which is to reduce the number of network parameters while maintaining its high accuracy on test data. A review is carried out of existing methods for pruning, which belong to different groups of approaches depending on their characteristics, such as dependence on input data and the need to consider network channels collectively. To solve this problem, approaches are proposed to compare network channels, based on the results of which parameters are selected to be removed. The approaches are based on the selection of an effective metric for assessing channel proximity and channel clustering. Pruning methods using the proposed approaches are described. The details of the software implementation of the methods are considered. The results of an experimental study of the efficiency of the proposed methods are presented.