Visual Transfer Between Atari Games Using Competitive Reinforcement Learning

[blopeur]

https://arxiv.org/abs/1809.00397

[AlphaWeaver]

Okay this is really cool! I know historically neural nets are hard to generalize to be applicable beyond one subject - using visuals to generalize is a simple idea... Inspired by how our brains work I guess, but it's cool that they got it to work!