Abstract

Generative models are very popular in a field of unsupervised learning.They are tremendously successful to learn underlying data distribution of training data and generate a new data with some variations.This paper presents a detailed study of generative models and how they differ from traditional discriminative models.The paper more focus on two most popular generative models such as Variational Autoencoder(VAE) and Generative Adversarial Network(GAN).The paper includes working of these generative models, their architecture and an experiment is conducted to generate images using very popular MNIST data set.The comparison between these two models and their advantages and disadvantages are presented based on an experiment. At last, some solutions are presented to further improve these models.

References

• Ian Goodfellow, M. Mirza, B. Xu, Y. Benjio. Generative Adversarial Network. Department of Computer Science and Research Operationl, University of Montreal(2014).
• Ian Goodfellow. NIPS 2016 Tutorial:Generative Adversarial Networks, from NIPS conference(2016).
• Diederik P. Kingma, Max Welling. Auto-Encoding Variational Bayes, Machine Learning Group, Universiteit van Amsterdam(2014).
• Adam Roberts, Jesse E., Douglas E. Hierarchical Variational Autoencoders for Music, Google Brain. from NIPS(2017).
• David Ha, Douglas E. A Neural Representation of Sketch Drawings, from arXiv:1704.03477v4 [cs.NE](2017).
• Alec Radford, Luke Metz,Soumith Chintala. UNSUPERVISED REPRESENTATION LEARNING WITH DEEP CONVOLUTIONAL GENERATIVE ADVERSARIAL NETWORKS, from ICLR(2016).
• Jun-Yan Z.,Taesung P.,Phillip I., Alexei E. Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks, from arXiv:1703.10593v5 [cs.CV](2018).
• Christan L.,Lucas T.,Ference H.Jose C.,Andrew C. Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network, arXiv:1609.04802v5 [cs.CV](2017).
• Han Z.,Tao X.,Hongsheng L.,Shaoting Z. StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks, arXiv:1612.03242v2 [cs.CV](2017).
• Hugo L.,Ole W.,Anders L.,Soren S. Autoencoding beyond pixels using a learned similarity metric, arXiv:1512.09300v2 [cs.LG](2016).
• Alireza M.,Brendan F.,Ian G. Adversarial Autoencoders, arXiv:1511.05644v2 [cs.LG](2016).
• Dimitris M.,Ian G.,Han Z. Self-Attention Generative Adversarial Networks, arXiv:1805.08318v1 [stat.ML](2018).
• Shane Barratt, Rishi Sharma. A Note on the Inception Score, arXiv:1801.01973v2 [stat.ML](2018).
• Martin H.,Hubert R.,Thomas U.,Bernhard N. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium, arXiv:1706.08500v6 [cs.LG](2018).