4 ms·
This is super cool, but it doesn't outperform deep learning based RL methods. In fact, I'm not sure how much more compute efficient than something like A3C it
by cshenton 9y ago
This is super cool, but it doesn't outperform deep learning based RL methods.
In fact, I'm not sure how much more compute efficient than something like A3C it would be. That can produce 4x the score of DQN in a comparable number of hours (and on a CPU).
- habitue 9y agoA3C is only ever run on one game at a time[0]. This paper gets good performance on all games with the same agent [0] read as: I have only seen papers with 1 agent per game for A3C
- cshenton 9y agoSo it can train on one game and play without training on a previously unseen (but also atari) game? That's pretty neat, DQN and A3C certainly can't do that.
- habitue 9y agoNo, in this case it is trained on all the games, but retains good scores on all of them. If you train basic AC3 in all the games, you'll get poor performance on all the games due to catastrophic forgetting