Self-play in game theory involves a player improving their strategy by playing against a copy of itself, while generalization involves applying knowledge from past experiences to new, similar situations. Both of these techniques rely on the concept of regret, and can be used for perfect information games.