Convolutional Neural Network Pruning: A Survey

Abstract

Deep convolutional neural networks have enabled remarkable progress over the last years on a variety of visual tasks, such as image recognition, speech recognition, and machine translation. These tasks contribute many to machine intelligence. However, developments of deep convolutional neural networks to a machine terminal remains challenging due to massive number of parameters and float operations that a typical model contains. Therefore, there is growing interest in convolutional neural network pruning. Existing work in this field of research can be categorized according to three dimensions: pruning method, training strategy, estimation criterion.