Skip to content Skip to navigation
Qualifying Exam
9/27/2018 11:00 am
Hill 482

Disconnected Manifold Learning for Generative Adversarial Networks

Mahyar Khayatkhoei, Dept. of Computer Science

Examination Committee: Prof. Ahmed Elgammal (Chair), Prof. Abdeslam Boularias, Prof. Pranjal Awasthi, Prof. William Steiger


Real images often lie on a union of disjoint manifolds rather than one  globally connected manifold, and this can hinder the training of common Generative Adversarial Networks (GANs). We first show that single  generator GANs are unable to correctly model a distribution supported on a disconnected manifold, and investigate the consequences. Next, we show  how using a collection of generators can address this problem. Finally,  we explain the serious issues caused by considering a fixed prior over  the collection of generators and propose a novel approach for learning  the prior and inferring the necessary number of generators without any  supervision.