Deep Neural Networks (DNNs) Fundamentals and Architectures
Mehdi Ghayoumi · 2021
This chapter reviews the deep learning (DL) models, concepts, and definitions. DL algorithms work with different data types. It processes data through units in different ordered sections by using different techniques in each layer. DL algorithms need a large volume of data for training, and the computation costs of making DL model is very high. One of the challenges in using DL is the vanishing gradient problem. As artificial neural networks, a DL network is better to cover generalization to increase the model flexibility. There are several methods like regularization to increase the generalization. The chapter presents an example for Deep Neural Networks with TensorFlow for the MNIST dataset. Convolutional neural network is one of the most popular deep learning algorithms with four main parts: convolution filters, pooling or subsampling, activation function, and fullyconnected layer. It shows promising results in visual classification problems.