The concept of probably approximately correct (PAC) learning has been foundational in computational learning theory.