In this seventh installment, we’re going to give you a taste of how great reinforcement learning is.
First, please check out the video below. We hope you can watch the 1:42 video without worrying about the English language.
You can see how Artificial Intelligence (AI) learns to break blocks intelligently through reinforcement learning.
（(When viewing with a smartphone, please turn your phone to the side. If you still can’t see it, please click here. You will be redirected to another company’s website (YouTube).
The awesomeness of reinforcement learning, contrasted with MAiMATE
Here’s a comparison between the video “Block Breaker” and “MAiMATE”.
|Prior knowledge before starting||Nothing; the AI doesn’t even know that I’m going to be a blockbreaker now.||Nothing; the AI doesn’t even know I’m about to make a trade.|
|How to start learning||Where the ball is released is random.||It’s random whether you go in from buying, selling, or doing nothing.|
|Learning Objectives||Make more score.||Make more profit.|
|The final system of evolution||The AI realizes that if it can pry open both ends and carry the ball upwards, it can mass-produce points with can-can-can scores, and it learns the technology to do so.||Technical and news information will tell us what to do when technical and news information is available, and we will find out what trading techniques the AI thinks are likely to win. The top rankings are considered to be very close to a situation where we are reaching the “two ends of the block breaker”.|
|When evolution does not go well||Because the game has a winning formula, you can expect to get to the winning formula as long as you take the time to learn it. In other words, it’s hard to assume if evolution doesn’t work.||Since there is no winning method in trading, it may be difficult to arrive at a method that you can be confident in. In that case, they become a reclusive type.|
This video was created by Google and I am presenting it as a case study in the application of reinforcement learning. (Thank you, Google.)
What do you all think? Isn’t this kind of exciting?
As far as I understand it, most services that say they are “using deep learning” are not “reinforcement learning”, but are being used for either “prediction” or “classification” purposes.
And I myself actually created an artificial intelligence (AI) that leveraged “deep learning” and did a series of forward tests, but unfortunately, I didn’t get results that gave me enough confidence to release the service, so I decided not to adopt it.
Nowadays, the trend of everything being based on artificial intelligence (AI) can be seen not only in Japan, but all over the world.
Whenever there is an opportunity to introduce technologies that have little to do with MAiMATE, we hope to provide good quality knowledge.