AI Podcast: Where Is Deep Learning Going Next?December 7, 2016
We’ll know AI really works when we hardly notice it at all, according to Bryan Catanzaro, a key figure in the field.
“AI gets better and better until it kind of disappears into the background,” says Catanzaro — NVIDIA’s head of applied deep learning research — in conversation with host Michael Copeland on this week’s edition of the new AI Podcast. “Once you stop noticing that it’s there because it works so well — that’s when it’s really landed.”
Bryan’s been in AI since the beginning. Or, as Michael says, as “about as long as it has really worked.” It’s a journey that’s taken him from UC Berkeley, where he earned his Ph.D., to NVIDIA, to Baidu — where he worked on a team that’s made a number of deep learning breakthroughs — and back to NVIDIA.
Along the way, he’s seen deep learning make incredible advances. It’s much further along than he would have predicted five years ago, he says. Image recognition has been one major success, with sophisticated facial recognition capabilities built into photo sharing services used by hundreds of millions of people every day.
“We’re at a point now where computers are actually better at recognizing objects in images than a person is,” Bryan says.
More’s coming, he explains. Deep learning — powered by ever more powerful GPUs — only grows more useful as the amount of data in the world grows.
“There’s a great number of problems that can be framed in this way, where you have a huge number of labeled examples, and you want a system to learn what that input means,” Bryan says.
To hear the whole conversation, tune into this week’s AI Podcast.
And if you missed our podcast last week, it’s definitely worth a listen: NVIDIA’s Will Ramey, a gifted explainer of all things deep learning, provides a clear explanation of the key concepts driving the field forward.