Generative Adversarial Networks Tutorial (article) - DataCamp

Generative Adversarial Networks Tutorial (article) - DataCamp







Good morning, and welcome to Go West, your weekly guide to life sciences news happening up and down the West Coast. I’m Rebecca Robbins, writing to you from my base in San Francisco.

I want to start this week's edition with an observation. In my recent conversations here with people in health and medicine, I keep hearing the same buzzword: GANs. At first I had no idea what it stood for — much less what it meant — so I nodded politely, and Googled later. GANs, it turns out, is short for "generative adversarial networks." It describes a machine learning technique that’s increasingly being used to interpret images.

Machine learning, of course, involves algorithms that incorporate feedback so that they get better over time at making decisions. Machine learning is itself a branch of artificial intelligence, that OG Silicon Valley buzzword.

So how does GANs work? The best analogy I found imagines a scenario in which a shop buys fine wine from sellers in order to resell it. Some of those sellers, however, are nefarious, and try to sell fake wine to the shop for profit. In response, the shop tries to detect those fake wines so it doesn’t make a bad purchase. Over time, the back-and-forth competition between the shop and the nefarious sellers makes both parties improve: The forgers learn better ways to disguise their fake wine, while the shop learns better ways of sniffing it out.

Which brings us back to GANs. The two machine learning systems that make up a GAN act in opposition to one another — with one trying to fool the other, which in turn tries to outwit its deceiver — until the whole process leads to better decisions. As GANs increasingly gets used in the life sciences, there’s hope that those better decisions could translate into better ways to read patients’ scans, record their data, and design their drugs.

You may have noticed, however, that hype is already rampant in AI and machine learning. It seems like a safe bet that GANs will get hyped, too, as the technique takes off in health and medicine. I’ll be watching closely.