New Learning Approach By Genetic Algorithm In A Convolutional Neural Network For Pattern RecognitionDownload PDF

29 Mar 2024 (modified: 21 Jul 2022)Submitted to ICLR 2017Readers: Everyone
Abstract: Almost all of the presented articles in the CNN are based on the error backpropagation algorithm and calculation of derivations of error, our innovative proposal refers to engaging TICA filters and NSGA-II genetic algorithms to train the LeNet-5 CNN network. Consequently, genetic algorithm updates the weights of LeNet-5 CNN network similar to chromosome update. In our approach the weights of LeNet-5 are obtained in two stages. The first is pre-training and the second is fine-tuning. As a result, our approach impacts in learning task.
TL;DR: Implement new approach without exerting backpropagation in learning of CNN is useful for parallel processing Like GPU.
Conflicts: .ir
Keywords: Deep learning, Supervised Learning, Optimization, Computer vision
21 Replies

Loading