Speed up neural network training with batch normalization

Sometimes training neural networks can take days or weeks. This has several disadvantages and has constantly troubled machine learning practitioners, especially those without access to more computing resources. Long training time often means that you cannot tune models until they are done training — several days or weeks. Even after they are done training, you cannot properly explore all possible tunings because of limited time to production. This is also not profitable for businesses as several man-hours are wasted whilst waiting for models to finish…