Artificial intelligence has never been as present — or as cool — as it is today. And, after years on the periphery, deep learning has become the most successful and most popular machine learning method around.
DL algorithms can now identify objects better than most humans, outperform doctors at diagnosing diseases, and beat grandmasters at their own board game. In the last year alone, Google DeepMind’s AlphaGo defeated one of the world’s greatest Go player — a feat most experts guessed would take another decade at least.
Some of the finest minds in AI are at the Re•Work Deep Learning Summit in London this week to discuss the entrenched challenges and emerging solutions to artificial intelligence through deep learning. Researchers from Google, Apple, Microsoft, Oxford, and Cambridge (to name a few) are in attendance or giving talks. Re•Work founder, Nikita Johnson told Digital Trends, “Our events bring together a multidisciplinary mix of three core communities: startups, academia, and industry, to encourage collaboration and discussion.”
Over the next few weeks we’ll explore these topics in depth and hear from experts about how intelligent algorithms will transform our everyday lives tomorrow and in the years to come.
But what exactly is deep learning?
Deep learning is a machine learning method that trains systems by using large amounts of data and multiple layers of processing.
Still confused? You’re not alone.
“People often say, ‘You can’t understand deep learning really. It’s too abstract,’” Neil Lawrence, professor of Machine Learning and Computational Biology at the University of Sheffield, quipped today during his opening presentation. “But I think people can grasp it intuitively.”
To help laymen — and even some enthusiasts — grasp the concept of deep learning, Lawrence drew a parallel to a classic carnival game, in which a player drops a ball down a pegged board to land it in a slot at the bottom. It’s a difficult task to reach a specific slot — almost purely chance. But imagine you could remove pegs to help guide balls in certain directions to designated slots. That’s something like to the task performed by deep learning algorithms.
“The difficult aspect is adjusting the ‘pegs’ such that the ‘yeses’ go into the ‘yes’ slot and the ‘nos’ go into the ‘no’ slot,” Lawrence said.
Sounds simple? It’s not.
It’s a problem people have grappled with for decades and it’s still far from solved. Even today’s best deep learning systems can do one task well but fail when they’re asked to do anything even marginally different. As DeepMind’s Raia Hadsell pointed out, you can spend weeks or months training an algorithm to play an Atari game but that knowledge can’t be generalized. In other words, you can teach a system to play Pong but have to start from scratch if you want to it play Space Invaders.
There may be solutions — Hadsell thinks her team at DeepMind has at least one — but the shortcomings show just how much work researchers have ahead of them.