REINFORCEMENT LEARNING

V diplomskem delu predstavljamo samoojačitveno učenje, ki je področje strojnega učenja in se ukvarja z vprašanjem, kako naj agent deluje v okolju, da doseže čim večjo nagrado. V nalogi opravimo splošen pregled te teme, nato podrobneje opišemo nekaj pomembnejših metod, eno izmed njih pa implementiramo v mrežnem okolju lovec-plen. Na koncu predstavimo še naš program ter analiziramo dobljene rezultate.In this diploma work we present reinforcement learning, which is an area of machine learning that studies the question of how an agent ought to act in an environment to achieve maximum reward. In this work we take a general look at the topic, then describe a few of the more important methods in detail and implement one of them in the predator-prey grid world domain. In the end, we present our program and analyze its results