Image generation and classification via generative adversarial networks

In recent years, bio image processing via supervised learning with convolutional neural networks (CNNs) has seen huge adoption in computer vision applications. Comparatively, unsupervised learning with CNNs has received less attention. Generative adversarial networks (GANs) are an emerging technique for both semi-supervised and unsupervised learning that has made a dramatic change in the computer vision field. GANs provide a way to learn deep representations without extensively annotated training data. They achieve this through deriving back propagation signals through a competitive process involving a pair of networks designed by game theory. GANs may be used in a variety of applications, including bio image generation, semantic image editing, style transfer, bio image super-resolution and also classification. Image classification is an issue that utilizes image processing, pattern recognition, and classification methods. We present two applications of GANs: classification, and the generation of images that humans find visually realistic. We focus on MGAN on the MNIST dataset and we apply this architecture in the form of distributed as our novelty. The proposed method experiment on the cluster with multiple machines to improve time-consuming with data distribution and removing the interaction between nodes during the training. Using the voting algorithm to increase the accuracy of classification. We prove the improvement by using T-Test. We conclude the proposed method decreases the training time and improves the accuracy by voting.