Bigger Models, Cleaner Splits and Human Baselines: Andrew Ng's Practical Recipe for Applied Deep Learning