AutoML: The Assembly Line of Machine Learning

Building real life machine learning applications need fair amount of tribal knowledge and intuition. Creating a machine learning model for a particular application requires many steps, usually manually performed by a data scientist or an engineer. These steps include ETL, feature engineering, model selection (including hyper-parameter tuning), safeguarding against data leakage, operationalizing models, scoring and updating models. 

Coupled with the explosion of ML use cases in the world that need to be addressed, there are not enough data scientists to build all the applications and democratize it.

Can we automate machine learning to make it truly accessible? After all, it’s not magic. It is math and statistics where we build models by generalizing examples. There is also some art to producing good models. In this session, you would learn about the building blocks of automated machine learning library Optimus Prime, built at Salesforce Einstein on top of Spark and Scala.