The process of deciding which experts in a mixture-of-experts model should process a given input token.