Koyejo wins prestigious Sloan Research Fellowship
CSL Assistant Professor Sanmi Koyejo recently earned a 2021 Sloan Research Fellowship from the Alfred P. Sloan Foundation. These early-career awards annually recognize 128 promising researchers from the United States and Canada who have the potential to make substantial contributions to their field.
Koyejo, computer sciences assistant professor will use the two-year $75,000 fellowship to further his research on improving the trustworthiness of machine learning (ML) systems when deployed in real-world systems. Specifically, he will continue to develop flexible ML tools that enhance the fairness, robustness, and privacy of algorithms used in healthcare and neuroimaging.
“This recognition from my peers is humbling and I’m excited to push my research agenda forward,” Koyejo said. “Of course, this research is possible because of the hard work of my excellent students, postdocs, and external collaborators.”
One key focus of Koyejo’s work revolves around ensuring that ML models transport properly from the training phase to real-world deployment. During the training phase, ML systems are often exposed to huge amounts of data, which enables them to identify patterns and make predictions similar to how humans learn and make decisions.
“It’s very common to have a machine learning system work well on training datasets, however when you apply the same system to real-world data, the system doesn’t work as well,” said Koyejo. “The hospital data you train [your system] on looks a little different to the hospital data that you’ll deploy to.”
Some of Koyejo’s most impactful work is being applied to a brain mapping project, where researchers investigate how different regions of the brain interact and give rise to specific human behaviors. The neuroscientists accumulate large amounts of data from multiple brain scans, however their measurement data may vary widely since it’s collected from different research centers using different imaging machines.
“This [variation] is an example of a transportability or trustworthiness problem in ML systems,” said Koyejo, whose underlying statistical models can make inferences about the unreliable data. “Some of the tools [my group] builds are designed to address these questions of data collection sensitivity and how to take advantage of data collected across several sites to make strong inferences.”
One of those tools—a probabilistic model with novel constraints—received a best paper award at the Conference on Uncertainty in Artificial Intelligence (UAI), a premier international meeting of research related to knowledge representation, learning, and reasoning in the presence of uncertainty.
Other aspects of his neuroscience work include foundational results that established graph-theoretical properties of dynamic brain connectivity, which may indicate a direct link between cognitive performance and the dynamic reorganization of the network structure of the brain. Koyejo’s brain encoding and decoding maps have been incorporated into Neurosyth, a publicly available online toolbox for brain data analysis.
Another application of Koyejo’s research tackles challenges related to using artificial intelligence and ML to diagnose COVID-19 and predict its severity by analyzing X-rays. One challenge to developing this technology is a lack of shared data across hospitals, often due to privacy and intellectual property concerns. Koyejo is developing privacy-preserving federated learning methods across a group of hospitals, including OSF HealthCare.
Koyejo is a founding member of Black in AI, an organization dedicated to increasing the number of Black people in the artificial intelligence field. He has received several other awards in recognition of his promising research—a Kavli Fellowship, a 2019 IJCAI Early Career Spotlight award, and a trainee award from the Organization for Human Brain Mapping.