Predicting disruptive instabilities in controlled fusion plasmas through deep learning