Jul 23, 2018 · 10 min

By Abe Haskins (Twitter, Github)

3.4K

6 responses

Jul 16, 2018 · 1 min

On “solving” Montezuma’s Revenge

2.5K

13

I think Montezuma’s Revenge is a bad problem — it is too difficult to be solved directly, yet simple enough to allow for workarounds such as learning by imitation… A bit like the Turing Test, which has only been solved using “tricks”.

It’s a game that I believe should be solved by “transfer”, making use of prior skills…

2

Jun 25, 2018 · 4 min

Dear friends,

I find myself here in Bordeaux, writing at an airy cafe and cinema space called Utopia. I accidentally ordered a ham, cheese, and stewed tomato sandwich for lunch, and so I paid my respects to the pig as I gingerly tasted ham for the first…

6

2 responses

Jan 9, 2018 · 1 min

Reinforcement learning (RL) practitioners have produced a number of excellent tutorials. Most, however, describe RL in terms of mathematical equations and abstract diagrams. We like to think of the field from a different perspective. RL…

4.1K

12 responses