Abstract: Supervised machine learning (SML) techniques, such as neural networks and gradient boosting machines, are being adopted in the banking industry due to their superior predictive performance and ability to do automated feature engineering. However, the black—box nature of these complex models poses challenges in regulated industries such as banking, where one needs to understand and explain the results to various stakeholders. Many interpretability tools have been (and are being) developed to address this challenge. In addition, attempts to build inherently interpretable machine learning model is becoming popular. This talk will start with an overview of the SML algorithms in banking and use some application to compare their performance. We will then describe techniques for machine learning interpretability and discuss some of our own research on post hoc methods, surrogate models, and inherently-interpretable models.

Jie Chen is Managing Director in Corporate Model Risk at Wells Fargo. She is currently managing the Cross-functional Model Validation team. She previously lead the Statistical Modeling and Machine Learning team in the Advanced Technologies for Modeling (AToM) Group for seven years, focusing on development of cutting-edge models, algorithms, and a computing platform to advance the Bank’s practice in the areas of credit, operational, and market risk management. She has over ten year experience on machine learning, artificial intelligence and advanced statistics in the banking industry. Jie holds a Ph.D. in Statistics from the Stewart School of Industrial and Systems Engineering at the Georgia Institute of Technology.

Linwei Hu is Vice President in the Advanced Technologies for Modeling (AToM) team inside Corporate Model Risk, Wells Fargo. He joined Wells Fargo in January 2015, shortly after he received his PhD degree in Statistics from University of Georgia. He has been with the bank ever since. His work is primarily on researching the latest statistics and machine learning methodologies, with a focus on interpreting black box machine learning models and developing self-interpretable machine learning models. In addition, he is also one of the contributor and maintainer of an internal python library for model validation.