Date of Publication :16th November 2017
Abstract: Generative Adversarial Networks or GANs were introduced by Ian Goodfellow and his colleagues at the university of Montreal. The concept behind these networks is that, two models fighting against each other would be able to co-train and eventually create a system that could learn more, with less help from humans, effectively reducing the huge amount of human effort required in training and creating deep learning models. GANs are the new class of two different deep neural networks which compete with one another to generate similar data, which leads to the creation of high quality fake information. What is the use of this generated data? If a computer can generate data, then it can use it to understand the scenario it is currently in. GANs are an interesting development in the domain of machine learning and more importantly, unsupervised learning. They can be applied in many fields, from generating text to predicting diseases. They are used for creating images from words, extracting high resolution images from low resolution pictures, currently research is also going on where it is used to generate new molecules which can be helpful in treating an ailment. They have been used also in games for generation of different scenes. Through this paper, we aim to understand what exactly Generative Adversarial Networks are and what are the existing applications of such models. We also consider the existing re-search challenges that exist in this domain and potential use cases that such a system could support
Reference :
-
- Improved Techniques for Training GANs, Tim Salimans, Ian Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, Xi Chen
- Chen, Xi, et al. ―Infogan: Interpretable representation learning by information maximizing generative adversarial nets.‖ Advances in Neural Information Processing Systems. 2016.
- I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, ―Generative adversarial nets,‖ in NIPS, 2014, pp. 2672–2680
- D. P. Kingma, S. Mohamed, D. J. Rezende, and M. Welling, ―Semi-supervised learning with deep generative models,‖ in NIPS, 2014, pp. 3581–3589.
- Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks, Alec Radford, Luke Metz, Soumith Chintala
- InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets. Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, Pieter Abbeel.
- A Learning Algorithm for Boltzmann Machines, David H. Ackley Geoffrey E. Hinton, Terrence J. Sejnowski
- Conditional Generative Adversarial Nets,Mehdi Mirza, Simon Osindero