Research Overview

My interests follow two research lines. 1) interpretable machine learning, where I love developing new forms of models that are understandable by humans, or reinventing classic models to incorporate modern considerations and desiderata for transparency, fairness, etc; 2) ML for the real-world problems. I’m fascinated by all kinds of real-world problems that can be solved by modern machine learning or reinforcement learning methods. To do that I will either use models I develop in research line 1) or customize new methods to solve a problem. I work on a wide range of topics covering business problems, healthcare, criminology, etc

Interpretable Machine Learning & Explainable AI

For this research line, I develop novel and generalizeable new methods in machine learning and reinforcement learning that develop human-understandable models or policies for prediction and decision making.

My recent focus has been on

  • decision rule-based models, for classification, regression, or causal analysis
  • hybrid models, where an interpretable model is constructed to compete and collaborated with any pre-trained black-box model, gaining model transparency at no or low-cost of predictive performance

    One paper on this topic was the finalist of the Best Paper Competition at 13th INFORMS Workshop on Data Mining & Decision Analytics Workshop, 2018.
    One paper on this topic was the runner-up of the Best Paper Competition at INFORMS Workshop on Data Science, 2019.
  • interpretable DNN, where a deep learning model is reinvented to incorporate careful design for higher interpretability.

Human-AI Collaboration

I work to develop AI models that work with humans in order to achieve more accurate and/or fair decisions..

     Our paper on this topic won the Best Paper Award at INFORMS Workshop on Data Science, 2020.

Machine Learning for Real-World Problems
I apply models developed above (generalizeable models) or customize some new methods for different real-world problems, covering a wide range of topics in healthcare, business problems, and criminal justice.

For example, my project on crime series detection has been cited by wikipedia and was a second-place winner for “Doing Good with Good OR” at INFORMS 2015.  Ideas from one of my project on crime series detection have been implemented by the NYPD (click here for link to article by Alex Chohlas-Wood and E.S. Levine) 

In 2018, we entered the FICO XML challenge and built a model to predict whether a customer’s loan application should be approved. We built a highly interpretable model which is accurate as any black-box model we trained. Our work won the FICO recognition award, 2018.

Research Overview was last modified: June 16th, 2021 by Tong Wang