A new study by researchers at the University of Toronto suggests that one of the fundamental assumptions of deep learning artificial intelligence models – that they require enormous amounts of training data to make accurate predictions – may not be a