Deep reinforcement learning overfits. Often, a trained network can only play the particular level(s) you trained it on! In our new paper, we show how to train more general networks with procedural level generation, generating progressively harder levels.https://t.co/VRtvIZ0OOz pic.twitter.com/JDNYPZuAHV
— Julian Togelius (@togelius) June 29, 2018
from Twitter https://twitter.com/togelius
June 29, 2018 at 08:57AM
via IFTTT
No comments:
Post a Comment