Latest in Culture

Image credit: Warner Bros.

Elon Musk's AI initiative opened an online dojo

The idea is to create standardized test environments for reinforcement-learning-based algorithms.
1 Shares
Share
Tweet
Share
Save
Warner Bros.

I'm going to go out on a limb here and say that the artificial intelligence you coded in your garage probably doesn't have the type of resources behind it that Google used to make DeepMind a fearsome Go competitor. That's what the Elon-Musk-backed OpenAI Gym is for. It's in open beta right now, and available test environments include Go on 9x9 and 19x19 boards, a ton of classic Atari games and robot control simulations, among others, with more to come.

The firm says it launched this is because progress made in reinforcement learning lags for a few reasons. Firstly, OpenAI notes that existing, open-source testing environments lack diversity and are difficult to set up and use. What's more, there's a dearth of standardization, which makes reproducing the tests -- key for any sort of academic research -- between different projects in an apples to apples way pretty hard to do.

Want in? The Python-based environments apparently are compatible with algorithms written in "any framework," including Tensorflow and Theano. But if you're looking to dominate the leaderboards you might be disappointed: OpenAI says it's worried less about your high scores and would rather see the generality of your technique when you upload your results.

From around the web

ear iconeye icontext filevr