Marzyeh Ghassemi’s journey into the realms of health and technology began in her childhood, rooted in her love for video games and puzzles. Growing up in a family with an engineering background in Texas and New Mexico, Ghassemi was encouraged to pursue STEM, leading to a profound fascination with health care and computer science. “While health care was a consideration, my passion for computer science and engineering ultimately prevailed,” Ghassemi states. She is now an associate professor at MIT’s Department of Electrical Engineering and Computer Science and the Institute for Medical Engineering and Science (IMES), also serving as the principal investigator at the Laboratory for Information and Decision Systems (LIDS).
The intersection of computer science and health care emerged as a significant focus for Ghassemi when she discovered the potential of artificial intelligence (AI) and machine learning (ML) in enhancing health outcomes. Today, her research group, Healthy ML, is dedicated to making machine learning systems more robust and equitable in health care applications.
In her formative years, Ghassemi was deeply influenced by her mother, who introduced her to advanced mathematical concepts, expanding her understanding beyond mere arithmetic. “Basic operations are important, but they can sometimes overshadow the idea that higher-level mathematics and sciences are fundamentally about logic and problem-solving,” she explains. Her mother’s support instilled a confidence in her that success in math could lead to fun and exciting challenges.
As Ghassemi progressed in her academic career, receiving her undergraduate degree from New Mexico State University, she encountered influential mentors who significantly impacted her path. Jason Ackelson, the director of the Honors College, aided Ghassemi in applying for a Marshall Scholarship, enabling her to study at Oxford University, where she earned her master’s degree in 2011. It was here that her interest in the rapidly advancing field of machine learning began to take shape. During her Ph.D. studies at MIT, she benefited from a collaborative environment filled with support from both professors and peers. Ghassemi emphasizes her commitment to replicating this nurturing atmosphere within her own research group.
While pursuing her Ph.D., Ghassemi made a critical discovery regarding biases within health data and their effects on machine learning models. After training various models to predict patient outcomes using health data, she realized that blindly utilizing all available data often overlooked significant disparities. During a pivotal meeting with Leo Celi, a principal research scientist at MIT, Ghassemi was prompted to analyze model performance across different genders, insurance types, and self-reported races, exposing existing gaps in efficacy. “Over almost a decade, we have shown that addressing these performance gaps is quite challenging, as they stem from intrinsic biases within health data and prevalent technical practices,” she notes. This revelation ignited her passion for exploring bias and fairness in ML applications.
A key breakthrough for Ghassemi involved demonstrating how learning models could identify a patient’s race from medical images—accomplishments that even trained radiologists struggled to achieve. However, the research group found that models designed to optimize general performance often performed poorly for women and minority groups. This summer, they combined their findings to reveal a concerning trend: the more a model learned to predict demographic characteristics from medical images, the more pronounced the performance gap became for those subgroups. The research team proposed that training models to accommodate demographic variances rather than concentrating solely on achieving overall average accuracy could mitigate these issues, although meticulous adjustments must be made at each deployment site.
“The research underscores that models optimized for one hospital setting may not translate optimally to another,” Ghassemi explains, stressing the implications for model developers and users. A model’s success in one environment does not ensure effectiveness elsewhere, highlighting a pressing need for a broader understanding of contextual applicability.
Ghassemi’s work is deeply intertwined with her identity as a visibly Muslim woman and a mother, shaping her worldview and influencing her research interests. “My focus on the robustness of ML models, coupled with the critical need to address existing biases, is no coincidence,” she asserts. She often finds inspiration in the outdoors, whether biking in New Mexico or walking along the Cambridge Esplanade, and approaches complicated problems by dismantling them into smaller components to understand where her assumptions may falter.
Though dedicated to her research, Ghassemi maintains a conscious effort to ensure her work does not define her entire identity. “It’s vital for academics to remain aware that our passions need not consume us,” she shares. “Cultivating interests and knowledge beyond one’s technical expertise is crucial.” She finds balance through meaningful relationships with family, friends, and colleagues who encourage her to embrace a fuller existence.
Recognized and awarded for her groundbreaking work at the intersection of computer science and health, Ghassemi views life as an evolving journey. Reflecting on a poignant quote from the Persian poet Rumi—“You are what you are looking for”—she emphasizes the importance of continual self-discovery and growth. “At every stage of life, we must reinvest in understanding who we are and intentionally guide ourselves toward who we aspire to be.”
In her multifaceted roles as a researcher, educator, and advocate for equitable health technologies, Marzyeh Ghassemi embodies the integration of passion and purpose, paving the way for innovative advancements in health care through machine learning.
Source link
