Maggie Makar receives NSF CAREER Award to develop machine learning models backed by causal reasoning

Maggie Makar, assistant professor of computer science and engineering at the University of Michigan, has received a National Science Foundation (NSF) CAREER Award, which will support her development of stronger artificial intelligence (AI) models based on imperfect data by leveraging causal mechanisms. Her work will contribute to the development of more robust and accurate predictive models for a wide range of applications, particularly in health care.

Given to a select number of early-career faculty across science and engineering fields, the NSF CAREER Award is a highly distinguished honor. The award aims to support young scientists and engineers with demonstrated excellence in research and teaching, and who “have the potential to serve as academic role models in research and education and to lead advances in the mission of their department or organization.”

Makar’s project, titled “From Fragile to Fortified: Harnessing Causal Reasoning for Trustworthy Machine Learning with Unreliable Data,” aims to leverage causal reasoning to build more accurate predictive models, particularly for unreliable or imperfect datasets. Although AI models have seen incredible advances in recent years, they still have several obstacles that limit their adoption in real-world settings.

One of these challenges is that machine learning models rely on large amounts of well-curated data to generate accurate predictions. When trained on imperfect or incomplete data, these models can learn to rely on spurious correlations and produce incorrect or misleading predictions. In a health care setting, in particular, this can have serious consequences.

To overcome these issues, Makar seeks to develop AI models that are able to generate accurate predictions based on smaller amounts of imperfect data. This will involve creating training schemes that prevent models from encoding correlations that are counter to known causal mechanisms, as well as building methods for quantifying uncertainty in model predictions based on potential gaps in causal knowledge.

Through these innovations, Makar plans to formulate robust and efficient machine learning models for application in clinical settings, particularly in the management of chronic pain. Although imperfect data in health contexts limits the utility of AI models, clinician knowledge about causal relationships can help fill in the gaps. Makar aims to leverage clinician expertise to build models that can perform well even when trained with incomplete data and can estimate the uncertainty of their predictions, better informing users’ decision-making.

With these stronger AI models, Makar hopes to add a valuable tool to clinicians’ arsenal and tackle a persistent medical challenge that affects millions.

Makar came to CSE as a Presidential Postdoctoral Fellow in 2021 before joining the faculty in 2023. She completed her PhD at MIT, where she worked in the Computer Science & Artificial Intelligence Laboratory (CSAIL). Her research interests lie at the intersection of machine learning and causal inference; she works to build robust machine learning models in resource-constrained settings based on causal mechanisms. Her work on this and related topics has been broadly published, with her papers appearing in top journals and conferences in the field.