English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Convolutional neural networks (CNNs) achieve state-of-the-art performance ina wide variety of tasks in computer vision. However, interpreting CNNs stillremains a challenge. This is mainly due to the large number of parameters inthese networks. Here, we investigate the role of compression and particularlypruning filters in the interpretation of CNNs. We exploit our recently-proposedgreedy structural compression scheme that prunes filters in a trained CNN. Inour compression, the filter importance index is defined as the classificationaccuracy reduction (CAR) of the network after pruning that filter. The filtersare then iteratively pruned based on the CAR index. We demonstrate theinterpretability of CAR-compressed CNNs by showing that our algorithm prunesfilters with visually redundant pattern selectivity. Specifically, we show theimportance of shape-selective filters for object recognition, as opposed tocolor-selective filters. Out of top 20 CAR-pruned filters in AlexNet, 17 ofthem in the first layer and 14 of them in the second layer are color-selectivefilters. Finally, we introduce a variant of our CAR importance index thatquantifies the importance of each image class to each CNN filter. We show thatthe most and the least important class labels present a meaningfulinterpretation of each filter that is consistent with the visualized patternselectivity of that filter.

Interpreting Convolutional Neural Networks Through Compression