Machine Learning Interpretability Toolkit

Abstract: With the recent popularity of machine learning algorithms such as neural networks and ensemble methods, etc., machine learning models become more like a ‘black box’, harder to understand and interpret. To gain the user’s trust, there is a strong need to develop tools and methodologies to help the user to understand and explain how predictions are made. Data scientists also need to have the necessary insights to learn how the model can be improved. Much research has gone into model interpretability and recently several open sources tools, including LIME, SHAP, and GAMs, etc., have been published on GitHub. In this talk, we present Microsoft's brand new Machine Learning Interpretability toolkit which incorporates the cutting-edge technologies developed by Microsoft and leverages proven third-party libraries. It creates a common API and data structure across the integrated libraries and integrates Azure Machine Learning services. Using this toolkit, data scientists can explain machine learning models using state-of-art technologies in an easy-to-use and scalable fashion.

Bio: Mehrnoosh Sameki is a technical program manager at Microsoft responsible for leading the product efforts on machine learning transparency within the Azure Machine Learning platform. Prior to Microsoft, she was a data scientist in an eCommerce company, Rue Gilt Groupe, incorporating data science and machine learning in retail space to drive revenue and enhance personalized shopping experiences of customers and prior to that, she completed a PhD degree in computer science at Boston University. In her spare time, she enjoys trying new food recipes, watching classic movies and documentaries, and reading about interior design and house decoration.