6

I'm currently a student doing some machine learning projects, and I want to use generative adversarial networks to train some data to discern for example, how old someone is. The intended output is a regression continuous output in term of months.

I see the DCGANs and GANs are targeting mainly classification problems. Can they be used to solve regression problems?

Thank you.

lppier
  • 161
  • 1
  • 3
  • GANs are not targettting classification problems. They are generative models, so their purpose is to generate data, usually images. – noe Aug 09 '18 at 08:02
  • @ncasas thanks, I just realized this, that the purpose of classifying was to get the generator's output as close as possible to the "real" images. – lppier Aug 09 '18 at 10:51
  • Semi-supervised Biomedical Translation with Cycle Wasserstein Regression GANs http://www.marzyehghassemi.com/wp-content/uploads/2018/01/semi-supervised-CWR-GAN_Ghassemi.pdf – user3710685 Jan 25 '19 at 16:43

3 Answers3

1

The answer is yes. There is a paper Intra-class Variation Isolation in Conditional GANs (https://arxiv.org/pdf/1811.11296.pdf) discussing it.

They propose a "C-GAN that is able to learn realistic models with continuous, semantically meaningful input parameters". They actually cover generating images of people at different ages as well.

user3023715
  • 203
  • 2
  • 5
1

It is possible to train a Convolutional regression model, and train it adversarially. So in theory you can. @Zhongsheng Chen linked a case of a Conditional GAN used for regression, even though I didn't find DCGANs for this, specifically.

In other words: it is possible in theory, but in practice using very large and heavy models such as GANs to run a regression is a like killing moquitoes with a bazooka.

Leevo
  • 6,225
  • 3
  • 16
  • 52
0

I am also interesting about it. Recently, I found a paper which the authors used a conditional GAN for regression tasks, in hope it will drive you over this problem. For a detail about this paper, see here.

Zhongsheng Chen
  • 11
  • 1