Computer Science Seminar: Chaofan Chen

This event is open to all Illinois Tech faculty and students. 

Abstract

As machine-learning models are playing increasingly important roles in many real-life scenarios, interpretability has become a key issue for whether we can trust the predictions made by these models, especially when we are making high-stakes decisions. Lack of transparency has long been a concern for predictive models in criminal justice and in health care. There have been growing calls for building interpretable, human-understandable machine-learning models, and “opening the black box” has become a debated issue in the media. Chaofan Chen's research addresses precisely the demand for interpretability and transparency in machine-learning models. The key problem of Chen's research is: “Can we build machine learning models that are both accurate and interpretable?”

To address this problem, Chen will discuss the notion of interpretability as it relates to machine learning, and present several new interpretable machine-learning models and methods he developed in his research. In particular, he will first give an overview of his research by discussing two types of model interpretability—predicate-based and case-based interpretability, and highlighting the contributions he has made. Chen will then focus on the topic of case-based interpretability for computer vision, in the remaining part of my talk. More specifically, I will present my work in developing deep neural networks that are able to reason about images by saying “this looks like that,” just like how we humans would explain to others on how to solve challenging image classification tasks. These networks are able to learn a meaningful latent embedding space that captures the notion of visual similarities and a set of prototypical cases for comparison. Given a new image, they are able to identify similar prototypical cases using distances in the latent space and make predictions according to the known class labels of those prototypical cases. The experiments on MNIST (for handwritten digit recognition) and CUB-200-2011 (for bird species identification) show that the case-based interpretable networks can achieve comparable accuracy with its analogous non-interpretable counterpart and at the same time, provide a level of interpretability that is absent in attention-based interpretable deep models. Indeed, as Chen's work has demonstrated, we can build machine learning models that are both accurate and interpretable by designing novel model architectures or regularization techniques.

Biography

Chen attended the University of Chicago and graduated with a Bachelor of Science in Mathematics (with honors). He began doctoral studies in computer science at Duke University in 2014. He was awarded the Outstanding Ph.D. Preliminary Exam Award in 2018 and the Outstanding Research Initiation Project Award in 2017 by the Department of Computer Science at Duke University. He pursued his research in the area of interpretable machine learning under the direction of Professor Cynthia Rudin.