Fascinated from infancy by the allure of video games and puzzles, Marzyeh Ghassemi’s interests were further piqued by a curiosity about health and wellness that emerged at an early stage. Luckily, she stumbled upon an opportunity to combine her two passions seamlessly.
“Although initially considering a career in healthcare, the allure of computer science and engineering proved irresistible,” notes Ghassemi, an affiliate professor at MIT’s Department of Electrical Engineering and Computer Science and Institute for Medical Engineering and Science, as well as principal investigator at the Laboratory for Information and Decision Systems. “When I stumbled upon the intersection of computer science and healthcare, specifically AI and ML, it was a thrilling convergence of interests.”
Currently, Ghassemi leads the Wholesome ML analysis group at LIDS, conducting groundbreaking research on leveraging machine learning (ML) to significantly strengthen its capabilities and ultimately improve healthcare outcomes by enhancing security and fairness.
Growing up in Texas and New Mexico within the culturally rich Iranian-American community that valued education and science, Ghassemi was influenced by her environment to pursue a career in a STEM field. As she delighted in puzzle-driven video games, finding solutions to unlock new levels or progress further presented a truly captivating challenge; meanwhile, her mother encouraged her to explore advanced math concepts at an early age, instilling in her the understanding that math transcended mere arithmetic.
According to Ghassemi, while emphasizing inclusion and multiplication is crucial, it’s essential not to overlook the fact that many advanced math and science concepts revolve around logic and problem-solving skills. Because of my mother’s inspiration, I realized that better times were ahead.
According to Ghassemi, she received significant support for her mental development from her mother, as well as numerous other individuals. As she received her undergraduate degree from New Mexico State University, Dr. Jason Ackelson, then director of the university’s Honors Faculty and a distinguished Marshall Scholar, served as inspiration for her future academic pursuits. The Division of Homeland Security facilitated her application for a Marshall Scholarship, which enabled her to pursue a master’s degree at Oxford University from 2008 to 2011; it was during this time that she developed a keen interest in the rapidly emerging field of machine learning. During her time at MIT, Ghassemi credits a supportive network with helping her achieve success, citing the assistance she received from both professors and peers. She aims to recreate this atmosphere of inclusivity and camaraderie in her own teaching practices, striving to foster an environment where her students feel valued and empowered to learn.
While pursuing her PhD, Dr. Ghassemi stumbled upon an initial indication that biases in healthcare data could be masked by machine learning models.
She developed fashion approaches to forecast outcomes by leveraging health data, as the prevailing mindset encouraged the integration of all available information. In image processing using neural networks, we observed that optimal features could be identified for superior performance, thereby obviating the need to manually craft specific features?
During a meeting with Leo Celi, principal analysis scientist at MIT’s Laboratory for Computational Physiology and IMERS, he asked whether Ghassemi had validated his models’ performance across patients from diverse gender, insurance, and self-reported racial backgrounds?
While Ghassemi conducted testing, inconsistencies still persisted. “For nearly a decade, research has consistently demonstrated that model disparities persist, stemming from entrenched biases in health data and outdated technical approaches.” Unless you take a closer look, fashions will unwittingly perpetuate and amplify biases, she suggests.
Throughout his career, Ghassemi has continued to delve into these matters, driven by an insatiable curiosity.
One of her most significant accomplishments was the result of several key factors coming together. Researchers found that analyzing fashion trends in medical images, such as chest X-rays, could inadvertently reveal a patient’s race, a distinction radiologists are currently incapable of making. The group found that fashion models optimized for general use underperformed when applied to women and minority groups. Last summer, they combined their findings to demonstrate that as a model learned to predict a patient’s race or gender from medical images, the greater its performance gap became for subgroups in these demographics? Ghassemi and her team found that the problem might be alleviated if a model were trained to accommodate demographic differences, rather than focusing solely on overall average performance – but this process would need to be repeated at every site where a model is deployed.
While fashions designed to maximize efficiency by balancing overall efficiency with the smallest possible equity gap may be optimal within a single hospital setting, they are unlikely to remain optimal in alternative settings. According to Ghassemi, this concept has a profound impact on the development of fashion designs for human consumption.
“One hospital may require the data to train a mannequin, subsequently verifying its performance under specific equity constraints.” Despite this, our analysis reveals that these efficiency gains do not persist in new environments? A mannequin well-suited to one environment may struggle to adapt in a distinct setting. The lack of transparency surrounding fashion’s influence on its own evolution poses a significant challenge, emphasizing the need for collective action to address this issue and ultimately benefit those responsible for designing and implementing fashionable innovations.
Ghassemi’s work is knowledgeable by her id.
“As a visibly Muslim woman and mother, these dual identities have profoundly shaped my perspective, influencing the lens through which I approach analysis and inquiry.” I investigate the resilience of machine learning models and explore how a lack of robustness can compound existing biases. The inquiry couldn’t possibly be an accidental occurrence.
As Ghassemi reflects on her creative process, she reveals that inspiration typically flows from time spent outdoors – whether cycling through the landscapes of New Mexico during her undergraduate days, rowing along Oxford’s waterways, conducting research as a PhD student at MIT, or simply taking leisurely strolls beside Cambridge’s picturesque Esplanade. When tackling complex issues, she finds that breaking down the problem into its constituent parts and questioning her assumptions about each component can be invaluable in uncovering potential pitfalls and blind spots?
“In her experience, the most significant barrier to innovation is often a misconception about one’s own knowledge.” “Uncovering the intricacies of a complex system often requires a meticulous approach, involving a thorough examination of models and components. Only by diving deep can one uncover hidden nuances and recognize the limitations of their initial understanding.”
As fervent as Ghassemi is about her profession, she thoughtfully preserves an awareness of life’s broader panorama?
As she notes, once an analyst becomes enamored with their findings, it can be challenging for them to disentangle those insights from their own identity – a pitfall many educators should strive to avoid. I strive to ensure that my pursuits and knowledge extend beyond the boundaries of my technical expertise.
“Prioritizing stability requires effective collaboration with high-performing team members.” Surround yourself with people who inspire and support your authenticity – hold onto those relationships!
With numerous accolades and widespread acclaim for his multifaceted contributions, encompassing both pioneering work in computer science and wellness initiatives, Ghassemi articulates a profound conviction that life is, indeed, a transformative journey.
“As she notes, the ancient wisdom of Persian poet Rumi is reflected in his translation: ‘You’re what you’re looking for.'” “At various junctures of one’s life, it is essential to reexamine and refine one’s sense of identity, guiding personal growth towards the individual they aspire to become.”
