Himabindu Lakkaraju

Timeline

1. 2011–2012

   Topic models and IBM Research

   Her master's thesis at the Indian Institute of Science in Bangalore develops semi-supervised topic models that extract sentiment and concepts from customer reviews, and the work wins the Best Research Paper Award at the 2011 SIAM International Conference on Data Mining. She then works for two years as a research engineer at IBM Research, India, in Bangalore and receives its Research Division Award in 2012.

2. 2016–2018

   Interpretable decision sets

   At Stanford University, she completes a computer science PhD advised by Jure Leskovec in 2018, collaborating with Jon Kleinberg, Cynthia Rudin and Sendhil Mullainathan. Her thesis work includes algorithms that automatically construct interpretable rules for classification, among them “Interpretable Decision Sets,” a 2016 paper at the ACM SIGKDD conference with Stephen Bach and Leskovec.

3. 2017

   Machine predictions for bail

   In February 2017, she coauthors “Human Decisions and Machine Predictions” with Jon Kleinberg, Jure Leskovec, Jens Ludwig and Sendhil Mullainathan, a study of how machine learning can inform judges' decisions on whether defendants await trial at home or in jail. Its policy simulation shows crime falling by up to 24.8% with no change in jailing rates. Microsoft Research awards her a 2017 dissertation grant for “Interpretable Machine Learning for Human Decision Making.”

4. 2018–2019

   Harvard postdoc and Innovators Under 35

   She joins Harvard University as a postdoctoral researcher in 2018. In June 2019, MIT Technology Review names her one of its Innovators Under 35 for an AI system that acts as a bias check for decision makers such as judges and doctors; schools in Montgomery County, Maryland, use her work to identify at-risk students. Vanity Fair names her one of the world's top innovators to watch in 2019.

5. November 2019

   Fooling LIME and SHAP

   In November 2019, she and Dylan Slack, Sophie Hilgard, Emily Jia and Sameer Singh show that LIME and SHAP, popular post hoc explanation methods that rely on input perturbations, are unreliable. Tested on real-world datasets including COMPAS, their scaffolding technique lets extremely biased classifiers fool both tools into producing innocuous explanations.

6. 2020

   Assistant professor at Harvard

   She becomes an assistant professor at Harvard Business School and Harvard's Department of Computer Science in 2020. The Trustworthy ML Initiative, which she co-founds in 2020, promotes research on the interpretability, fairness, privacy and robustness of machine learning models and gives newcomers easy access to resources.

7. 2022–2025

   OpenXAI and a Sloan Fellowship

   With her collaborators, she releases OpenXAI in 2022, an open-source framework for evaluating and benchmarking post hoc explanation methods. In 2023, she receives an NSF CAREER Award and is named a Kavli Fellow by the National Academy of Sciences. The Indian Institute of Science gives her its Distinguished Young Alumnus Award in 2024, and she becomes a Sloan Research Fellow in computer science in 2025.

8. As of October 2026

   Harvard professor and Google scientist

   She is an assistant professor at Harvard University with appointments in Harvard Business School and the Department of Computer Science, and a part-time senior staff research scientist at Google. Her research focuses on AI safety, interpretability and reliability, with an emphasis on foundation models; she leads the AI4LIFE research group at Harvard and is principal investigator of the Trustworthy AI Lab at the Harvard Business School AI Institute.