In Search of a Common Deep Learning Stack