Why Are Generative Adversarial Networks Vital for Deep Neural Networks? A Case Study on COVID-19 Chest X-Ray Images
The need to generate large scale datasets from a limited number of determined data is highly required. Deep neural networks (DNN) is one of the most important and effective tools in machine learning (ML) that required large scale datasets. Recently, gener
- PDF / 1,214,771 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 15 Downloads / 213 Views
M. Y. Shams (B) Faculty of Artificial Intelligence, Kafrelsheikh University, Kafr Elsheikh 33511, Egypt e-mail: [email protected] O. M. Elzeki Faculty of Computers and Information, Mansoura University, Mansoura 35516, Egypt e-mail: [email protected] M. Abd Elfattah Misr Higher Institute for Commerce and Computers, MET, Mansoura, Egypt e-mail: [email protected] A. E. Hassanien Faculty of Computers and Artificial Intelligence, Cairo University, Giza, Egypt e-mail: [email protected] T. Medhat Department of Electrical Engineering, Faculty of Engineering, Kafrelsheikh University, Kafr Elsheikh 33516, Egypt e-mail: [email protected] © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2020 A.-E. Hassanien et al. (eds.), Big Data Analytics and Artificial Intelligence Against COVID-19: Innovation Vision and Approach, Studies in Big Data 78, https://doi.org/10.1007/978-3-030-55258-9_9
147
148
M. Y. Shams et al.
Keywords Deep neural network · Generative adversarial network · Machine learning stochastic gradient descent (SGD) · Improved adam COVID-19 X-ray chest
1 Introduction Deep Neural Networks (DNN) is inspired by the human biological brain consisting of neurons, synapses, and much more. Artificial neural networks (ANN) are key to building a DNN because they consist of multiple hidden layers stacked. DNN was formulated from hierarchical neural networks to improve the process of classifying supervised patterns [1]. In the process of training DNN, transfer learning is an effective and powerful tool to enable the training of large-scale datasets without over-fitting problem results from the target dataset that is much smaller than the basic dataset [2]. There are many attempts to formulate DNN, for example, multilayer perceptron (MLP) as well as backpropagation that consists of feed-forward and feedback ANN [3]. DNN can be utilized as a feature extractor and classifier as well. However, to learn one layer of DNN feature vectors at a time, the multiple layers of feature vectors can be used as a starting point for a discriminative that is called the “fine-tuning” phase during which backpropagation through the DNN slightly adjusts the weights found in pre-training [4]. In this chapter, we utilized two different optimizers during the training step for generating COVID-19 X-Ray chest images based on GAN architecture. The first optimizer was the stochastic gradient descent (SGD), and the second was the improved Adam optimizer (IAdam) [5]. The loss function is determined for the two applied optimizers with minimum loss values. The main contribution of this chapter is to enlarge limited datasets to produce augmented COVID-19 X-Ray images with a minimum loss function. On the other hand, we prove the ability of GAN architecture using two both SGD and IAdam optimizers. The rest of this chapter is organized as background in Sect. 2, which survey using DNN in COVID-19 X-Ray images detection and classifications. Section 3 demonstrates that
Data Loading...
