Wayne Roseberry, Microsoft
Sure, automation is great because it can give you a whole bunch of information really fast. And yea, it is a big pain because it gives you a whole bunch of information really fast. Sometimes the amount is too much to process, but you know you need to look because there might just be an important failure in the pile that has to be fixed.
What if it was possible to predict your decision before you read it? How could that save time? How could that help your team focus on important work first rather than dig through a huge pile of data.
This paper will talk about experiments the author has executed in the Microsoft Office team using machine learning to predict outcomes for failures found by massive automation suites. It will talk about the methodology, the problems involved with evaluating such results, the potential discovered in the experiments and some of the inherent challenges, difficulties and risks in trusting such decisions to a machine.
Target Audience: Intermediate