Applied Machine Learning and AI for Engineers

Chapter 14. Azure Cognitive Services

As a child growing up in the 1960s, I idolized the Apollo astronauts. Swaggering out to the launch pad and riding flame-breathing rockets into space, they were my superheroes. But the group I really wanted to emulate—the people I wanted to be—were the engineers in Mission Control. Seated in front of their CRT screens in white shirts and black ties, chatting with the astronauts and poised to spring into action at the first sign of trouble, they were the epitome of cool. They used computers less powerful than today’s smartphones to put men on the moon—a scientific achievement that is unsurpassed to this day.

Thanks to deep learning, computers today can perform feats of magic that the engineers in Mission Control could only have dreamed of. They can recognize objects in images, translate text and speech to other languages, identify people in video feeds, turn art into words and words into art, and more. But state-of-the-art deep-learning models are too complex—and too costly—for the average engineer or software developer to build. Microsoft reportedly spent hundreds of thousands of dollars training the ResNet model that won the 2015 ImageNet Large Scale Visual Recognition Challenge. Creating that model required a great deal of expertise, massive amounts of GPU time, and millions of images.

A welcome trend in AI today is AI as a service. Microsoft, Amazon, Google, and other tech giants employ professional data scientists who build sophisticated ...