stat946w18/AmbientGAN: Generative Models from Lossy Measurements: Difference between revisions

From statwiki
Jump to navigation Jump to search
m (Basic information)
Line 1: Line 1:
https://openreview.net/pdf?id=Hy7fDog0b
= Introduction =
Generative Adversarial Networks operate by simulating complex distributions but training them requires access to large amounts of high quality data. Ashish Bora, Eric Price and Alexandros G. Dimakis propose AmbientGAN as a way to recover the true underlying distribution from lossy data under certain conditions. Even when these conditions are not met the results from AmbientGAN still produce high quality results.
 
= Problem Specification =
 
= Model =
AmbientGAN compared to GAN (AmbientGAN passes the generator output through a measurement function)
 
= Results =
 
= Open questions =
 
= Criticisms =
 
= References =
https://openreview.net/forum?id=Hy7fDog0b

Revision as of 03:08, 26 February 2018

Introduction

Generative Adversarial Networks operate by simulating complex distributions but training them requires access to large amounts of high quality data. Ashish Bora, Eric Price and Alexandros G. Dimakis propose AmbientGAN as a way to recover the true underlying distribution from lossy data under certain conditions. Even when these conditions are not met the results from AmbientGAN still produce high quality results.

Problem Specification

Model

AmbientGAN compared to GAN (AmbientGAN passes the generator output through a measurement function)

Results

Open questions

Criticisms

References

https://openreview.net/forum?id=Hy7fDog0b