Netflix Is 'Training' Its Recommendation System By Using Amazon's Cloud To Mimic The Human Brain

Advertisement

reed hastings netflix

Getty Images/ Gareth Cattermole

Frequent Netflix users rejoice: Netflix is working on a new technology that should make its recommendation engine much better.

Advertisement

The goal of the technology is to stop recommending movies based on what you've seen, and instead make suggestions based on what you actually like about your favorite shows and movies.

Right now, Netflix looks at the things you watch and based on stuff like the actors, genre, filming location, it offers suggestions.

That method is far from perfect. Sometimes Netflix gets stuck in rut and doesn't offer enough variety. Just because you like "Parks and Recreation" doesn't automatically mean you like "The Office" and "30 Rock" and nothing else.

And it can miss the subtle differences between similar shows - the thing that makes a person love "The West Wing" may not apply to "House of Cards."

Advertisement

That's why Netflix is moving into a field of research known as "deep learning." That means that Netflix is "training" its software to provide better recommendations by feeding massive amounts of information to a tech called "neural networks." Neural networks mimic how the human brain identifies patterns.

The company took the lessons learned by researchers at Google, Stanford, and Nvidia and created deep learning software that takes advantage of Amazon's powerful cloud infrastructure, according to a new post on Netflix's technology blog.

It's no surprise that Netflix is building its neural network tech on top of Amazon's cloud, as it's one of the largest customers of Amazon Web Services.

In this case, Netflix used Amazon servers built with processors from Nvidia. These processors are typically used for processing effects in video games, photos or other graphics. With its human-brain software running on Amazon's computers, Netflix could "train" a new machine in 47 minutes, compared to the 20 hours of previous efforts.

With times like that, Netflix's engineers can set up a training session, have it run, and see results in the same day - and then run another test with improvements based on those results.

Advertisement

That doesn't mean recommendations are going to start getting better every hour of the day. What it does mean is that the humans running it will be able perform more tests and get insight into patterns than they never could have identified before.

Netflix isn't the only player in tech trying to utilize deep learning to improve its service. Back in December, Facebook hired NYU professor Yann LeCun, an expert in the field of "unsupervised learning," a field in AI research that focuses on making computers that can teach themselves. IBM has its Watson supercomputer - which the company has said will soon be able to reason and debate - and Google shares a quantum AI lab with NASA.