Neil Jethani

60 5th Ave

New York, NY 10012

I am an MD/PhD candidate at the NYU Grossman School of Medicine. My work focuses on building machine learning tools for healthcare. Specifically, I am interested in developing methods that help explain machine learning predictions to physicians and expand the diagnostic utility of the ECG. I am fortunate to be advised by Dr. Rajesh Ranganath and Dr. Yindalon Aphinyanaphongs.

Before graduate school, I earned a BS from the University of California, San Diego in Bioengineering. As an undergraduate student, I worked on projects investigating biomaterials for biomimetic systems and analyzing sequencing data.

In general, I try to have an active life outside of research. I enjoy doing anything that puts me in nature: hiking, biking, rock climbing, surfing, swimming, etc.

news

Aug 31, 2021	New work introducing a method that estimates Shapley values to provide model explanations in real-time with a single forward pass is added to arXiv.
Feb 25, 2021	Our work on developing a new machine learning explanation method and explanation evaluation method is accepted to AISTATS 2021.
Oct 22, 2020	Our work on characterizing myocardial injury in patients hospitalized With COVID-19 is accepted to Circulation.

selected publications

Have We Learned to Explain?: How Interpretability Methods Can Learn to Encode Predictions in their Interpretations

Jethani, Neil, Sudarshan, Mukund, Aphinyanaphongs, Yindalon, and Ranganath, Rajesh

2021

Abs arXiv

While the need for interpretable machine learning has been established, many common approaches are slow, lack fidelity, or hard to evaluate. Amortized explanation methods reduce the cost of providing interpretations by learning a global selector model that returns feature importances for a single instance of data. The selector model is trained to optimize the fidelity of the interpretations, as evaluated by a predictor model for the target. Popular methods learn the selector and predictor model in concert, which we show allows predictions to be encoded within interpretations. We introduce EVAL-X as a method to quantitatively evaluate interpretations and REAL-X as an amortized explanation method, which learn a predictor model that approximates the true data generating distribution given any subset of the input. We show EVAL-X can detect when predictions are encoded in interpretations and show the advantages of REAL-X through quantitative and radiologist evaluation.