Himabindu Lakkaraju
Harvard University, MA·Computer Science·Early-career
Algorithms increasingly influence consequential choices about people: who receives a loan, which job applicants advance, which treatment a patient is offered. Because these models are complex and often proprietary, the people affected cannot see why a decision came out the way it did.
- Fields
Computer Science
Mathematics & Statistics
- Topics
explainable AI
interpretable machine learning
model explanations
- Grants
- 1 known grant, through Jul 2028
What they work on
- Build theoretical and empirical frameworks that measure how faithfully leading explanation methods capture the behavior of different model classes, including linear and non-linear machine learning models.
- Conduct large-scale user studies with domain experts in healthcare, lending and hiring to quantify whether existing explanations are useful for real high-stakes decisions.
- Train algorithmic agents on the expert-study data so they imitate expert judgment, then use the agents to evaluate explanation methods at scale.
- Design algorithms that automatically match an appropriate explanation method to a given model and real-world application context.
More about the research
Dozens of "explanation methods" now exist to describe model behavior in human terms, but they rest on different assumptions and frequently disagree. Without a way to judge them, practitioners cannot tell which explanation to believe.
Why it matters
Teams deploying predictive models in hospitals, banks and hiring pipelines currently choose among conflicting explanation tools with little evidence. Benchmarks and selection algorithms from this work could help them pick methods that reliably reflect what a model is doing, giving doctors, loan applicants and job candidates a more trustworthy account of automated decisions that affect them.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Xueru Zhang
Ohio State University, OH·Computer Science
Model how individuals strategically and repeatedly change their behavior in response to fair machine learning decisions, using a sequential framework that combines stochastic control and game theory.
Xueru Zhang
Ohio State University, OH·Computer Science
Build a mathematical model of how individuals strategically and repeatedly adapt to machine learning decision systems, and how this feeds back into model training. The model combines game theory and stochastic control.
Juba Ziani
Georgia Institute of Technology, GA·Computer Science·Early-career
Model how individual data providers shape the outcomes of machine learning systems through strategic choices about what data to share, combining tools from computer science and economics.
The school
Harvard University: 42 labs on PI Pages, $1.5B a year on research, 712 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.