Output details

A Pong playing agent modelled with massively overlapping cell assemblies

<24> In this work, a neural agent learns to play the pong video game. It learns by imitating a human user, and by itself with feedback via failure, in under five minutes of simulated and real time. It works by learning to associate ball positions with paddle positions. Failure is compensated for by the ball activating the portion of the net where the paddle should be when the system fails. The novel learning of overlapping cell assemblies enable the system to generalise rapidly as similar ball positions require similar paddle positions.