Remove to Improve?

Kamila Abdiyeva, Martin Lukac, Narendra Ahuja

Research output: Chapter in Book/Report/Conference proceedingConference contribution


The workhorses of CNNs are its filters, located at different layers and tuned to different features. Their responses are combined using weights obtained via network training. Training is aimed at optimal results for the entire training data, e.g., highest average classification accuracy. In this paper, we are interested in extending the current understanding of the roles played by the filters, their mutual interactions, and their relationship to classification accuracy. This is motivated by observations that the classification accuracy for some classes increases, instead of decreasing when some filters are pruned from a CNN. We are interested in experimentally addressing the following question: Under what conditions does filter pruning increase classification accuracy? We show that improvement of classification accuracy occurs for certain classes. These classes are placed during learning into a space (spanned by filter usage) populated with semantically related neighbors. The neighborhood structure of such classes is however sparse enough so that during pruning, the resulting compression bringing all classes together brings sample data closer together and thus increases the accuracy of classification.

Original languageEnglish (US)
Title of host publicationPattern Recognition. ICPR International Workshops and Challenges, 2021, Proceedings
EditorsAlberto Del Bimbo, Rita Cucchiara, Stan Sclaroff, Giovanni Maria Farinella, Tao Mei, Marco Bertini, Hugo Jair Escalante, Roberto Vezzani
Number of pages16
ISBN (Print)9783030687953
StatePublished - 2021
Event25th International Conference on Pattern Recognition Workshops, ICPR 2020 - Milan, Italy
Duration: Jan 10 2021Jan 11 2021

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume12663 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349


Conference25th International Conference on Pattern Recognition Workshops, ICPR 2020

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science


Dive into the research topics of 'Remove to Improve?'. Together they form a unique fingerprint.

Cite this