Corpus ID: 46373531

Reducing Duplicate Filters in Deep Neural Networks

@inproceedings{RoyChowdhury2018ReducingDF,
  title={Reducing Duplicate Filters in Deep Neural Networks},
  author={Aruni RoyChowdhury and Prakhar Sharma and E. Learned-Miller},
  year={2018}
}

Aruni RoyChowdhury, Prakhar Sharma, E. Learned-Miller

Published 2018

people.cs.umass.edu

Save to Library

Create Alert

Cite

Launch Research Feed

This paper investigates the presence of duplicate neurons or filters in neural networks. This phenomenon is prevalent in networks and increases with the number of filters in a layer. We observe the emergence of duplicate filters over training iterations, study the factors that affect their concentration and compare existing network reducing operations. We validate our findings using convolutional and fully-connected networks on the CIFAR-10 dataset.