Ian Goodfellow, inventor of the generative adversarial network (GAN), is scary smart.
From the Verge: Apple has poached another of Google’s top AI researchers:
At 34, Goodfellow is young to be an AI researcher with so much clout. But he’s well known for his work inventing a type of AI system known as a generative adversarial network or GAN.
GANs are two-part networks, formed of both a generator and a discriminator. The generator is trained on a dataset and then tries to replicate it. It shows its work to the discriminator which attempts to distinguish between the training dataset and the new output. If it can tell the difference, it sends the generator back to the drawing board to try again.
This sounds abstract, but GANs have proved to be wonderfully productive little systems. They’re particularly good at generating fake photos, videos, audio, and text. Those AI-generated portraits of imaginary people? GANs. That software that turns your doodles into photorealistic paintings? GANs again.
My take: Goodfellow’s first implementation of GAN was coded in one sitting, starting at midnight, after he sketched out the idea at a bar. A guy like that is worth 1,000 engineers.
He was interviewed by Andrew Ng, one of his Stanford mentors, in 2017. Cue the 15-minute YouTube: