A related concept to an “episode” is the set of everything that mediates the agents’ achievement of reward, i.e. it’s every X that lies on a path A⇢X⇢R for some action A and reward R. These “mediators” X carve out a part of both time and space. An episode is roughly the convex hull of those mediators.
A related concept to an “episode” is the set of everything that mediates the agents’ achievement of reward, i.e. it’s every X that lies on a path A⇢X⇢R for some action A and reward R. These “mediators” X carve out a part of both time and space. An episode is roughly the convex hull of those mediators.