Interpretable Machine Learning


With widespread use of machine learning, there have been serious societal consequences from using black box models for high-stakes decisions, including flawed bail and parole decisions in criminal justice, racially-biased models in healthcare, and inexplicable loan decisions in finance. Transparency and interpretability of machine learning models is critical in high stakes decisions. However, there are clear reasons why organizations might use black box models instead: it is easier to profit from inexplicable predictive models than transparent models, and it is actually much easier to construct complicated models than interpretable models. Most importantly, there is a widely-held belief that that more accurate models must be more complicated, and more complicated models cannot possibly be understood by humans. Both parts of this last argument, however, are lacking in scientific evidence and are often not true in practice. There are many cases in which interpretable models are just as accurate as their black box counterparts on the same dataset, as long as one is willing to search carefully for such models.

In her talk, Dr. Rudin will discuss the interesting phenomenon that interpretable machine learning models are often as accurate as their black box counterparts, giving examples of such cases encountered throughout her career. One example she will discuss is predicting manhole fires and explosions in New York City, working with the power company. This was the project that ultimately drew Dr. Rudin to the topic of interpretable machine learning. This project was extremely difficult due to the complexity of the data, and interpretability was essential to her team’s ability to troubleshoot the model. In a second example, she will discuss how interpretable machine learning models can be used for extremely high stakes decisions, such as caring for critically ill patients in intensive care units of hospitals. Here, interpretable machine learning is used to predict seizures in patients being monitored with continuous electroencephalogram monitoring (cEEG). In a third example, she will discuss predicting criminal recidivism, touching upon the scandal surrounding the use of a black box model in the U.S. justice system, questioning whether we truly need such a model at all.

About the speaker

Cynthia Rudin Cynthia Rudin is a professor of computer science, electrical and computer engineering, statistical science, and biostatistics & bioinformatics at Duke University, and directs the Interpretable Machine Learning Lab (formerly the Prediction Analysis Lab). Previously, Prof. Rudin held positions at MIT, Columbia, and NYU. She holds an undergraduate degree from the University at Buffalo, and a PhD from Princeton University. She is a three-time winner of the INFORMS Innovative Applications in Analytics Award, was named as one of the “Top 40 Under 40” by Poets and Quants in 2015, and was named by as one of the 12 most impressive professors at MIT in 2015. She is a fellow of the American Statistical Association and a fellow of the Institute of Mathematical Statistics.