| Title |
Hybrid Pruning of Deep Learning System |
| Authors |
이성준(Sungjun Lee) ; 서기성(Kisung Seo) |
| DOI |
https://doi.org/10.5370/KIEE.2020.69.11.1750 |
| Keywords |
Deep learning; Pruning; Convolutional neural network; Filter reduction; Genetic algorithm; APoZ |
| Abstract |
A convolutional neural network (CNN), which is one of the deep learning models, has been very successful in a variety of computer vision tasks. Filters of a CNN are automatically generated, however, they can be further optimized since there exist the possibility of existing redundant and less important features. Evolutionary algorithms is adopted to remove the unnecessary filters in order to minimize the parameters of CNN networks while maintaining a good performance of classification. Previous approach using evolutionary algorithm shows excellent performance for the filter reduction, but has a limitation of huge amount of computation.
Hybrid approach combining evolutionary algorithm and weight based pruning (APoZ) is proposed to enhance the efficiency of reduction and to reduce computation complexity. We demonstrate the proposed filter reduction method performing experiments on CIFAR-10 data based on the classification performance. |