In his proposal, “Deriving Robust Interpretation for Clinical Risk Prediction Models,” Dr. Wang details his plan to create better-understood clinical risk prediction models. One roadblock the associate professor of population health sciences has seen with these “black boxes” is that feeding models with different training data produces different algorithms. Ultimately, the interpretations of these models are going to be different. This results in uncertainty over how a model came to a certain conclusion – and if that conclusion is really the best one.
The Google Faculty Research Award provides a year’s worth of funding to world-class faculty pursuing cutting-edge research, along with tuition for one graduate student. This will allow Dr. Wang and his selected student to make these complicated models more consistent and robust.
That also means increasing inclusivity, an initiative Dr. Wang is passionate about.
“One big thing about application of AI models in healthcare is bias. If you train a model based on a white population, it is unlikely it is going to work well for a black population,” he said. “If you want a model to be stable, you need it to be inclusive. If we do this project well, we could help solve these inclusivity problems.”
Although Dr. Wang and his student will be focused on improving predictive risk models in healthcare, their research could be used across a wide variety of fields. “Models that are not stable and robust are deemed essentially useless,” he explains. Dr. Wang sees this as a fairly broad problem that hasn’t been studied widely, yet.
“This is not just for healthcare,” Dr. Wang said. “People care about how to interpret – in manufacturing, finance, wherever you build models to make predictions. But they haven’t thought too much about how stable or robust the interpretation could be.”
While the award provides resources for a year, Dr. Wang sees this research leading to long-term future collaborations. Google has provided a jumpstart to get the research off the ground.
“Google’s Faculty Research Award is particularly competitive because of the reputation of the company,” Dr. Wang said. “They reviewed over 900 submissions globally that are strong, from good schools and good researchers. It’s definitely a pleasure.”