An improved data stream summary - The count-min sketch and its applications.pdf - arXiv:1705.07831v1[cs.LG 22 May 2017 Stabilizing GAN Training with

An improved data stream summary - The count-min sketch and its applications.pdf

This preview shows page 1 - 2 out of 17 pages.

Stabilizing GAN Training with Multiple Random Projections Behnam Neyshabur Srinadh Bhojanapalli Ayan Chakrabarti Toyota Technological Institute at Chicago 6045 S. Kenwood Ave., Chicago, IL 60637 { bneyshabur,srinadh,ayanc } Abstract Training generative adversarial networks is unstable in high-dimensions when the true data distribution lies on a lower-dimensional manifold. The discriminator is then easily able to separate nearly all generated samples leaving the generator without meaningful gradients. We propose training a single generator simultane- ously against an array of discriminators, each of which looks at a different random low-dimensional projection of the data. We show that individual discriminators then provide stable gradients to the generator, and that the generator learns to produce samples consistent with the full data distribution to satisfy all discrimi- nators. We demonstrate the practical utility of this approach experimentally, and show that it is able to produce image samples with higher quality than traditional training with a single discriminator. 1 Introduction Generative adversarial networks (GANs), introduced by [ 1 ], endow neural networks with the ability to express distributional outputs. The framework includes a generator network that is tasked with producing samples from some target distribution, given as input a (typically low dimensional) noise vector drawn from a simple known distribution, and possibly conditional side information. The generator learns to generate such samples, not by directly looking at the data, but through adversarial training with a discriminator network that seeks to differentiate real data from those generated by the generator. To satisfy the objective of “fooling” the discriminator, the generator eventually learns to produce samples with statistics that match those of real data. In regression tasks where the true output is ambiguous, GANs provide a means to simply produce an output that is plausible (with a single sample), or to explicitly model that ambiguity (through multiple samples). In the latter case, they provide an attractive alternative to fitting distributions to parametric forms during training, and employing expensive sampling techniques at the test time. In particular, conditional variants of GANs have shown to be useful for tasks such as in-painting [ 2 ], and super-resolution [ 3 ]. Recently, [ 4 ] demonstrated that GANs can be used to produce plausi- ble mappings between a variety of domains—including sketches and photographs, maps and aerial views, segmentation masks and images, etc . GANs have also found uses as a means of un-supervised learning, with latent noise vectors and hidden-layer activations of the discriminators proving to be useful features for various tasks [ 2 , 5 , 6 ].
Image of page 1
Image of page 2

You've reached the end of your free preview.

Want to read all 17 pages?

  • Fall '05
  • RudolfFleischer

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask You can ask ( soon) You can ask (will expire )
Answers in as fast as 15 minutes