Goodfellow, Bengio, GAN, Generative Adversarial Nets

<p>Oct 02, 2016 05:41 PM &nbsp;background paper &nbsp;from Judy Hoffman for Vision Reading Group

(Mondays at 4:30PM in 463A Gates.)</p>

<p>

minimax 2 player game... analogy...</p>

<p>

In the proposed adversarial nets framework, the generative model is pitted against an adversary: a discriminative model that learns to determine whether <b>a sample is from the model distribution </b>or the <b>data distribution.</b>&nbsp;</p>

<p>

&nbsp;

The generative model can be thought of as analogous to a<b> team of counterfeiters, trying to produce fake currency and use it without detection,</b>&nbsp;</p>

<p>

&nbsp;

while the discriminative model is analogous to the<b> police, trying to detect the counterfeit currency.</b> Competition in this game drives both teams to improve their methods until the counterfeits are indistiguishable from the genuine

</p>