Captivated as a toddler by video games and puzzles, Marzyeh Ghassemi was also fascinated at an early age in health. Luckily, she found a path where she could mix the 2 interests.
“Although I had considered a profession in health care, the pull of computer science and engineering was stronger,” says Ghassemi, an associate professor in MIT’s Department of Electrical Engineering and Computer Science and the Institute for Medical Engineering and Science (IMES) and principal investigator on the Laboratory for Information and Decision Systems (LIDS). “After I found that computer science broadly, and AI/ML specifically, may very well be applied to health care, it was a convergence of interests.”
Today, Ghassemi and her Healthy ML research group at LIDS work on the deep study of how machine learning (ML) might be made more robust, and be subsequently applied to enhance safety and equity in health.
Growing up in Texas and Latest Mexico in an engineering-oriented Iranian-American family, Ghassemi had role models to follow right into a STEM profession. While she loved puzzle-based video games — “Solving puzzles to unlock other levels or progress further was a really attractive challenge” — her mother also engaged her in more advanced math early on, enticing her toward seeing math as greater than arithmetic.
“Adding or multiplying are basic skills emphasized for good reason, but the main focus can obscure the concept much of higher-level math and science are more about logic and puzzles,” Ghassemi says. “Due to my mom’s encouragement, I knew there have been fun things ahead.”
Ghassemi says that along with her mother, many others supported her mental development. As she earned her undergraduate degree at Latest Mexico State University, the director of the Honors College and a former Marshall Scholar — Jason Ackelson, now a senior advisor to the U.S. Department of Homeland Security — helped her to use for a Marshall Scholarship that took her to Oxford University, where she earned a master’s degree in 2011 and first became eager about the brand new and rapidly evolving field of machine learning. During her PhD work at MIT, Ghassemi says she received support “from professors and peers alike,” adding, “That environment of openness and acceptance is something I try to duplicate for my students.”
While working on her PhD, Ghassemi also encountered her first clue that biases in health data can hide in machine learning models.
She had trained models to predict outcomes using health data, “and the mindset on the time was to make use of all available data. In neural networks for images, we had seen that the suitable features could be learned for good performance, eliminating the necessity to hand-engineer specific features.”
During a gathering with Leo Celi, principal research scientist on the MIT Laboratory for Computational Physiology and IMES and a member of Ghassemi’s thesis committee, Celi asked if Ghassemi had checked how well the models performed on patients of various genders, insurance types, and self-reported races.
Ghassemi did check, and there have been gaps. “We now have almost a decade of labor showing that these model gaps are hard to deal with — they stem from existing biases in health data and default technical practices. Unless you consider carefully about them, models will naively reproduce and extend biases,” she says.
Ghassemi has been exploring such issues ever since.
Her favorite breakthrough within the work she has done got here about in several parts. First, she and her research group showed that learning models could recognize a patient’s race from medical images like chest X-rays, which radiologists are unable to do. The group then found that models optimized to perform well “on average” didn’t perform as well for ladies and minorities. This past summer, her group combined these findings to show that the more a model learned to predict a patient’s race or gender from a medical image, the more severe its performance gap could be for subgroups in those demographics. Ghassemi and her team found that the issue may very well be mitigated if a model was trained to account for demographic differences, as a substitute of being focused on overall average performance — but this process needs to be performed at every site where a model is deployed.
“We’re emphasizing that models trained to optimize performance (balancing overall performance with lowest fairness gap) in a single hospital setting should not optimal in other settings. This has a vital impact on how models are developed for human use,” Ghassemi says. “One hospital may need the resources to coach a model, after which have the ability to show that it performs well, possibly even with specific fairness constraints. Nevertheless, our research shows that these performance guarantees don’t hold in recent settings. A model that’s well-balanced in a single site may not function effectively in a distinct environment. This impacts the utility of models in practice, and it’s essential that we work to deal with this issue for individuals who develop and deploy models.”
Ghassemi’s work is informed by her identity.
“I’m a visibly Muslim woman and a mother — each have helped to shape how I see the world, which informs my research interests,” she says. “I work on the robustness of machine learning models, and the way a scarcity of robustness can mix with existing biases. That interest just isn’t a coincidence.”
Regarding her thought process, Ghassemi says inspiration often strikes when she is outdoors — bike-riding in Latest Mexico as an undergraduate, rowing at Oxford, running as a PhD student at MIT, and nowadays walking by the Cambridge Esplanade. She also says she has found it helpful when approaching a sophisticated problem to think concerning the parts of the larger problem and take a look at to know how her assumptions about each part could be incorrect.
“In my experience, essentially the most limiting factor for brand spanking new solutions is what you think that you understand,” she says. “Sometimes it’s hard to get past your personal (partial) knowledge about something until you dig really deeply right into a model, system, etc., and realize that you just didn’t understand a subpart accurately or fully.”
As passionate as Ghassemi is about her work, she intentionally keeps track of life’s greater picture.
“Once you love your research, it might be hard to stop that from becoming your identity — it’s something that I feel a whole lot of academics have to concentrate on,” she says. “I attempt to be sure that that I actually have interests (and knowledge) beyond my very own technical expertise.
“Among the best ways to assist prioritize a balance is with good people. If you might have family, friends, or colleagues who encourage you to be a full person, hold on to them!”
Having won many awards and far recognition for the work that encompasses two early passions — computer science and health — Ghassemi professes a faith in seeing life as a journey.
“There’s a quote by the Persian poet Rumi that’s translated as, ‘You might be what you’re on the lookout for,’” she says. “At every stage of your life, you might have to reinvest find who you’re, and nudging that towards who you must be.”
