Unsupervised feature learning-based encoder and adversarial networks

Abstract In this paper, we propose a novel deep learning-based feature learning architecture for object classification. Conventionally, methods are trained with supervised But, would require large amount of training data. Currently there increasing trends to employ unsupervised learning. By doing so, dependency on the availability data could be reduced. One implementation is where network designed “learn” features automatically from obtain good representation that then used Autoencoder and generative adversarial networks (GAN) examples methods. For GAN however, trajectories may go unpredicted directions due random initialization, making it unsuitable To overcome this, hybrid encoder convolutional (DCGAN) architectures, variant GAN, proposed. Encoder put top Generator avoid initialisation. We called our method as EGAN. The output EGAN two neural (DCNNs): AlexNet DenseNet. evaluate proposed three types dataset results indicate better performances achieved by compared using autoencoder GAN.