Drilling and Building: The Power Apps of Machine Learning

Patterns are what machine-learning algorithms exist to sniff out. But detecting those patterns is almost never the endgame. Typically, we use machine learning (a category in which I also include deep learning) to drill down to the patterns most relevant to some decision-support scenario, such as identifying fine-grained nuances of customer sentiment for use in target marketing or pinpointing the signs of imminent equipment failure through continuous sifting of scattered event-log databases. Once discovered, the statistical patterns can take on a programmatic life of their own that goes far beyond decision support in potential applications. As crystallized in machine-learning models, the patterns can become key assets in the development of other algorithmic applications that have little or no relevance to decision support.