Structural Compression of Convolutional Neural Networks
This addresses the interpretability challenge in machine vision for researchers and practitioners, though it is incremental as it builds on existing pruning methods.
The authors tackled the problem of convolutional neural networks (CNNs) being difficult to interpret due to millions of weights, by introducing CAR, a greedy structural compression scheme that prunes filters to obtain smaller and more interpretable CNNs while achieving close to original accuracy, with compressed networks retaining filter diversity with an order of magnitude less filters.
Deep convolutional neural networks (CNNs) have been successful in many tasks in machine vision, however, millions of weights in the form of thousands of convolutional filters in CNNs makes them difficult for human intepretation or understanding in science. In this article, we introduce CAR, a greedy structural compression scheme to obtain smaller and more interpretable CNNs, while achieving close to original accuracy. The compression is based on pruning filters with the least contribution to the classification accuracy. We demonstrate the interpretability of CAR-compressed CNNs by showing that our algorithm prunes filters with visually redundant functionalities such as color filters. These compressed networks are easier to interpret because they retain the filter diversity of uncompressed networks with order of magnitude less filters. Finally, a variant of CAR is introduced to quantify the importance of each image category to each CNN filter. Specifically, the most and the least important class labels are shown to be meaningful interpretations of each filter.