If you’re a regular reader of this blog, you’ll remember that I’ve talked about Generative Adversarial Networks (GANs) before.
And you may recall that I also mentioned Robbie Barrat, a very cool Machine Learning researcher who uses GANs to create stunning works of computer-generated art.
GANs are made up of two competing neural networks: a generator (a deconvolutional network) that tries to create new works of art, and a discriminator (a CNN) that tries to detect which art works were created by the generator.
We start with a pre-trained discriminator that can recognize art of a specific style, for example 18th century portraits.
Then we slowly improve the generator.
At first, the discriminator will detect and block every work of art created by the generator.
But eventually, the generator will become smart enough to create stunning new works that fool the discriminator every time.
Here are some works that Robbie created with a GAN that has been trained on portraits:
All of Robbie’s code is open source. You can download it from here:
The French art collective Obvious also downloaded Robbie’s GAN and trained it on a new set of portraits.
This is the portrait their fully-trained GAN eventually created:
Then, in a very smart publicity move, they offered the portrait to Christy’s for auction.
They expected to get about $10,000 for it.
But that’s not what happened.
During the auction, an anonymous bidder offered to pay $432,000 for the portrait!
Bit overpriced if you ask me 😉
But my point is: this is one of many untapped opportunities in Machine Learning.
Obvious correctly assumed that people are willing to buy computer-generated paintings. So they built an app to test their theory, they created a painting, offered it to Christy’s, and they won big.
There are many other opportunities for success still out there.
And the right idea could make you rich.
How are you going to win big?
Do you have a cool idea?