To build the feature libraries for auto-featurization, we leverage algorithms from decades of Microsoft research in natural language processing, machine learning, computer vision, speech, big data and much more – the same algorithms that power products such as Bing, Cortana and Microsoft Office. Today we have started with a basic set of text featurizers, and we will be continually expanding the selection overtime. For example, coming soon, we will be adding support for deep neural network based featurizers. The power of these feature libraries is that they are usually trained on a large amount of data that is not available to most users (e.g. DNN image featurizer, trained on tens of millions of annotated images, or DSSM, trained on years of click data from Bing Ads and web search), and they save users weeks or months relative to training their own complex models.