The algorithm designed by researchers from University of Toronto in Canada outperformed its own training by 9%