World Journal of Engineering Research and Technology

Oleg Namicheishvili, *Mikheil Ramazashvili, Zhuzhuna Gogiashvili and Natia Namicheishvili

ABSTRACT

From automatic image annotation to voice recognition, from computers capable of defeating champions in Go to autonomous vehicles, many recent successes in artificial intelligence are based on deep neural networks. Deep learning has become a key topic in discussions today.Artificial neural networks are essentially non-parametric, potentiallycomplex models: unlike linear regression, they allow for the creation of highly flexible models.In this article, we discuss the fundamental principles of multilayer perceptrons and their training. We will also briefly touch on deep (hierarchical) neural networks. The objectives are:— Understanding the decision-making process of single-layer and multilayer perceptrons;— Implementing the training procedure of a perceptron;— Explaining the update mechanism of a multilayer perceptron using backpropagation;— Understanding the key principles of building and training a multilayer perceptron;— Understanding the tasks of deep learning.

[Full Text Article] [Download Certificate]