Friday, June 23, 2017

Almost everything you need to know about Reinforcement Learning in 1 minute and 12 seconds

This article by at titled:

Atari-Playing AI Learns From Your Mistakes

Ms. Mlot brings out a number of key a very interesting points:
"After some training, the AI managed to outperform humans at pinball and match them at Q*bert and Space Invaders. But the computer struggled with Ms. Pacman and Montezuma’s Revenge (named after an expression for diarrhea contracted while visiting Mexico).

“Atari has recently become a benchmark for testing reinforcement learning algorithms,” Kurin said, citing a 2015 experiment in which Google’s DeepMind created an artificial intelligence platform that taught itself how to play 50 Atari titles.

Read more about the Atari Grand Challenge Dataset in a paper published last week by Kurin, fellow RWTH Aachen University researchers Lucas Beyer and Bastian Leibe, and Sebastian Nowozin and Katja Hofmann of the Machine Intelligence and Perception Group at Microsoft Research in Cambridge."

The video below really drives home the point of reinforcement learning:

No comments:

Post a Comment