Implicit Filter Sparsification In Convolutional Neural Networks
This incremental finding addresses the problem of model compression for efficient deployment in deep learning applications.
The paper demonstrates that implicit filter-level sparsity naturally occurs in CNNs using Batch Normalization, ReLU, adaptive gradient descent, and L2 regularization, achieving feature sparsity comparable to or better than explicit pruning methods.
We show implicit filter level sparsity manifests in convolutional neural networks (CNNs) which employ Batch Normalization and ReLU activation, and are trained with adaptive gradient descent techniques and L2 regularization or weight decay. Through an extensive empirical study (Mehta et al., 2019) we hypothesize the mechanism behind the sparsification process, and find surprising links to certain filter sparsification heuristics proposed in literature. Emergence of, and the subsequent pruning of selective features is observed to be one of the contributing mechanisms, leading to feature sparsity at par or better than certain explicit sparsification / pruning approaches. In this workshop article we summarize our findings, and point out corollaries of selective-featurepenalization which could also be employed as heuristics for filter pruning