At least in an RL context, relies on transferring learned representations from one task to a similar one. Disadvantage: When the representation changes considerably from game to game.