Abstract:

This article discusses the architecture of Convolutional Neural Networks in detail. All primary layers and their hyperparameters were described and presented in depth. The main idea of training neural networks with the inclusion of weights and other parameters and updating them based on backward propagation was described. The influence and role of the loss function and the optimizer in the neural network are discussed and the most frequently used types of loss functions and optimizers are listed and characterized. The second part of the article presents an overview of deep learning frameworks and descriptions of their most commonly used types. Furthermore, examples of using two of most popular frameworks were presented.