A Survey on Filter Pruning Techniques for Optimization of Deep Neural Networks

Deep Neural Networks (DNNs) have been an important and fast-developing tool used for computer vision, and artificial intelligence. Since these algorithms are widely used for image classification, they are bound to a few issues, creating a need for the DNN models to be optimized. The need for optimization is created due to computational complexity, the number of parameters and model size. Pruning techniques have been employed to mitigate this issue in DNNs, one of these techniques is Filter pruning. There are huge numbers of methods under Filter pruning that have been proposed and each one of them is based on specific sub-objectives. In this paper, we aim to represent different types of pruning methods in a summarized way and conclude on a method that is most efficient in delivering pruned model. The conclusion is stated after trying the methods in a common environment of data set and computational system.