Transcript of a Two Minute Papers YouTube episode (Two Minute Papers #72) discussing reinforcement learning and OpenAI's Gym. Speakers explain reinforcement learning basics (actions, rewards, long-term planning challenges), note Google DeepMind's interest in strategy games, mention curiosity-driven approaches to aid long-term planning, and introduce OpenAI Gym as a unified framework with environments, reference implementations, and leaderboards to standardize testing and comparison of RL methods.
Chapters
Dear fellow scholars, this is two minute.
Papers Withai Reinforcement learning is a technique in the field of machine learning to learn how to navigate in a labyrinth, play a video game, or to teach a digital creature to walk.
Usually we are interested in a series of actions that are in some sense.
Optimal in a given environment.
Despite the fact that many enormous tomes.
Exist to discuss the mathematical details, the intuition behind the algorithm itself is incredibly simple. Choose an action and if you get.
Rewarded for it, keep doing it.
If the rewards are not coming, try something else.
The reward can be, for instance, our score in a computer game or how far our digital creature could walk.
It is usually quite difficult to learn things where the reward comes long after.
Our action, because we don't know when.
Exactly the point was when we did something well.
This is one of the reasons why Google DeepMind will try to conquer strategy games in the future, because this is a genre where good plays usually include long term planning that reinforcement learning techniques.
Don'T really excel at.
By the way, this just in.
They have just published an excellent paper.
On including curiosity in this equation in.
A way that helps long term planning.
Remarkably, as more techniques pop up in.
This direction, it is getting abundantly clear that we need a framework where they can undergo stringent testing.
This means that the amount of collected.
Rewards and scores should be computed the.
Same way and in the same physical framework.
OpenAI is a non profit company boasting.
An impressive roster of top tier researchers.
Who embarked on the quest to develop open and ethical artificial intelligence techniques.
We've had a previous episode on this.
When the company was freshly founded and as you might have guessed, the link is available in the description box. They have recently published their first major project that goes by the name Jim.
Gym is a unified framework that puts reinforcement learning techniques on an equal footing.
Anyone can submit their solutions which are run on the same problems and as.
A nice bit of gamification, leaderboards are.
Established to see which technique emerges victorious.
These environments range from a variety of.
Computer games to different balancing tasks.
Some simpler reference solutions are also provided for many of them as a starting point.
This place is like Disney World for.
Someone who is excited about the field of reinforcement learning.
With more and more techniques, this subfield gets more saturated. It gets more and more difficult to.
Be the first at something. That's a great challenge for researchers.
From a consumer point of view, this.
Means that better techniques will pop up.
Day by day and as I like.
To say quite often we have really.
Exciting times ahead of us. A quick shout out to experiment a startup to help research projects come to fruition by crowdsourcing them. Current experiments include really cool projects like how we could implement better anti doping policies for professional sports or how to show on a computer screen how our visual imagination works. Thanks for watching and for your generous support, and I'll see you next time.
