Probably Approximately Correct (PAC) learning analyzes machine learning mathematically using probability bounds.