top | item 10790286

I Trained a Deep Q Network Built in TensorFlow to Play Atari Pong

39 points| superfx | 10 years ago |reddit.com

5 comments

order
[+] smhx|10 years ago|reply
This is simply a transpile / reproduction of the original Torch version from Deepmind, but in TensorFlow. It doesn't really do anything new or different compared to the paper by [Mnih et. al.](http://www.nature.com/nature/journal/v518/n7540/full/nature1...) .
[+] tostitos1979|10 years ago|reply
Replicating a previous result is still a worthwhile endeavor. Frankly, I don't think the adjective "simply" applies here. This isn't a script kiddie running a binary they found on the web.

P.S. I'm a professional scientist and I think the world would be better if people replicated results more often, and even showcased negative results.

[+] nullc|10 years ago|reply
Would be interesting to see if it could be regularized to make it a bit less twitchy. (E.g. by giving a fitness bonus to no action / and-or penalizing many changes of direction within some time window.)