Redes neurais recorrentes lstm e suas aplicações

Abstract

The objective of this work is to review the Long Short-Term Memory (LSTM) artificial neural network model and its application in different activities, in order to evaluate its versatility. Four activities are presented, one of which is the recognition and classification of digits in real time and other three of data generation, namely: production of text fragments, prediction of pollution indexes and realization of movements in the Pong game. Four models of the LSTM recurrent neural network were developed, being implemented in Python and trained through the algorithm of backpropagation through time. The network proved to be robust in all proposed scenarios and extremely flexible, since few adaptations were necessary in the neural network model to be applied to a different application. Based on the obtained results, it can be concluded that there is a great potential of use of the network in several applications.