S-lstm-gan: shared recurrent neural networks with adversarial training

Abstract

In this paper, we propose a new architecture Shared-LSTM Generative Adversarial Network (S-LSTM-GAN) that works on recurrent neural networks (RNNs) via an adversarial process and we apply it by training it on the handwritten digit database. We have successfully trained the network for the generator task of handwritten digit generation and the discriminator task of its classification. We demonstrate the potential of this architecture through conditional and quantifiable evaluation of its generated samples.