Student Seminar Series
DATE: Tuesday, October 31st, 2017, 4:00pm
LOCATION: MSB 1147 (Colloquium Room).
SPEAKERS: Jilei Yang, Graduate Student, Statistics, UC Davis
TITLE: “Gaining interpretation for black-box predictive models”
ABSTRACT: Despite high predictive performance, most of machine learning models lack interpretability. However, feature reasoning is very important for experts with domain knowledge to understand and interpret the model and gain insights for further improvements. In this talk, I will explore a new model interpretation and feature reasoning approach for arbitrary predictive models. I will first introduce two concepts for feature reasoning, namely, feature contribution and feature influence. Then I will show how to obtain feature contribution and feature influence for each individual sample in the dataset. I will also propose three evaluation criteria on how well this approach interprets a model. Finally, I will demonstrate this approach by using customer churn/upsell dataset. This project is part of my work as a data scientist intern at LinkedIn.
This seminar series is organized by PhD Students Irene Kim, Ozan Sonmez and Clark Fitzgerald.