Way back to she will be able to keep in mind, Cathy Wu ’12, MNG ’13 needed to search out methods to resolve issues to enhance folks’s lives. Her mother and father have been Taiwanese immigrants, and her father had an extended commute to his job, which took him away from the household. On a good price range, the remainder of the household usually stayed house on a road that was too busy for enjoying open air. Wu and her siblings ended up enjoying quite a lot of pc video games.
Wu says her want to make the world a greater place, her dad’s every day battle in opposition to site visitors, and the video games she performed, like “SimCity,” have been the seeds of her motivation to design secure, environment friendly transportation programs.
Wu is an affiliate professor within the MIT Division of Civil and Environmental Engineering (CEE) and the Institute for Knowledge, Programs, and Society (IDSS), and a principal investigator within the Laboratory for Info and Choice Programs. Her analysis focuses on utilizing machine studying and reinforcement studying (RL) to advance dependable methods for bettering a spread of advanced programs, together with transportation.
“Designing transportation programs consists of modeling and analyzing dozens, if not a whole lot or 1000’s, of variants, which implies that an evidence-driven method to designing these programs is just not inside attain of at the moment’s instruments,” Wu says. “That is the position that RL performs. If profitable, it could free transportation researchers and allow their practitioner companions to design the programs they need.”
Wu credit her older sister with instilling in her the need to enhance folks’s lives, and Wu’s curiosity in transportation suits neatly into that preferrred.
“I like transportation as a result of it connects everybody. All of us use it, all of us expertise it, all of us have points with it. So, at some degree, we’re all within the system being higher,” she says.
Wu acquired inquisitive about making use of synthetic intelligence to transportation whereas incomes her undergraduate diploma at MIT, after attending a lecture on autonomous autos by the late professor Seth Teller. The lecture, which Teller gave throughout an Impartial Actions Interval robotics competitors (that Wu truly gained), was the occasion that honed her specific method to transportation analysis, Wu says. She started working with Teller, and when he stopped concentrating on autonomous autos, he inspired Wu to switch to Professor Daniela Rus, who had finished analysis on robotaxis.
“I’m very grateful to the individuals who helped me discover these pursuits and helped me turn out to be the particular person I’m now,” she says, particularly naming Teller, Rus, and “my associates at Dropbox,” who invited her to do a second internship centered on transportation points.
After her grasp’s diploma at MIT, Wu went on to earn her PhD on the College of California at Berkeley. Throughout that point, she noticed that transportation researchers have been spending years creating optimization strategies to mannequin and analyze a single new variant of a system. Her method as a pc scientist working to develop RL and optimization methodologies to handle transportation challenges held the promise of exponentially improved effectivity.
In 2018, Wu’s final yr of her PhD at UC Berkeley, she efficiently utilized RL to a site visitors downside: mechanically analyzing the potential site visitors circulation influence of autonomous autos in a spread of various site visitors networks. The analysis went viral.
Whereas this might have been a “the remaining is historical past” second for Wu, RL turned out to be a flighty pal. Wu labored on RL concept in a postdoc at Microsoft and got here again to MIT as school drawn, she says, by the sustainability focus of CEE, and IDSS’s emphasis on infusing knowledge science into different disciplines.
But over the following two years, Wu’s additional makes an attempt to use RL to site visitors issues failed.
“That was annoying,” Wu says, “it was unclear whether or not the issue was me (the advisor), my college students, the site visitors area, or RL itself.”
Nonetheless, the sooner analysis was a proof-of-concept demonstration that RL could possibly be utilized to transportation programs.
And in 2022, she and her college students recognized that RL algorithms are so delicate that an algorithm that works on one downside could not on even a intently associated one. A key consequence, which Wu says she is proudest of “as a result of it was like the sunshine on the finish of an extended tunnel of damaging outcomes,” got here in 2023. She and her staff of researchers devised a option to work across the sensitivity of RL. The staff discovered that whereas RL could not prepare nicely on 90 p.c of a bunch of issues, it could possibly prepare fairly nicely on 10 p.c. And by coaching RL fashions on these issues that remedy and generalize nicely, the resultant fashions collectively carry out nicely on a set of associated issues, even people who wouldn’t have been solved by means of direct coaching. The researchers designed an algorithm to find out which issues to make use of RL to coach, and that algorithm improved coaching effectivity by as much as 30 instances, which means that what would usually have required 100 coaching fashions could solely require three fashions.
“This work gave me again the boldness that reinforcement studying can play an essential position in fixing arduous optimization issues, together with in transportation,” Wu says. “Now, a great chunk of my group works on the subject of contextual RL, which is the setting the place RL seeks to resolve an area of associated issues.”
Wu’s newer analysis applies RL to resolve a tough transportation optimization downside with essential coverage implications: the work exhibits that eco-driving measures through which car speeds are intelligently managed to scale back extreme stopping and beginning might scale back car emissions by between 11 and 22 p.c. The system gives proof that insurance policies instituting such measures might considerably enhance system effectivity, and is “an indication that RL can be utilized to tell transportation coverage on issues of sensible significance,” Wu says.
“I’m a giant fan of evidence-based coverage and consider it’s the idea for a thriving democratic society, but our societal programs are so advanced,” Wu says. “Individuals can bicker perpetually about what’s higher or worse, however I do consider that there are questions we bicker about that may be analyzed systematically utilizing knowledge and have goal solutions. A big a part of the explanation I’m in academia is to raised perceive how know-how can help democratic societal decision-making.”
Wu says that a lot of the work she and her staff have finished over the past a number of years has produced algorithms “to streamline the event of solvers for arduous optimization issues, whether or not they’re associated to transportation or to different programs, similar to logistics, provide chains, manufacturing, and useful resource allocation.
“This alludes to my most well-liked type of labor,” Wu says, “which is known as use-inspired fundamental analysis,” explaining that such analysis addresses a sensible downside, creating elementary information that always interprets to different sensible issues. Her college students begin by probing consequential issues starting from security to congestion to accessibility, figuring out the place present strategies fall quick, and permitting the issues themselves to form the path of the analysis.
On the similar time, Wu’s want to assist others on a extra private degree performs out in her educating.
“I really like working with college students, each within the classroom and analysis mentoring,” she says. “It makes my day when I’m able to educate somebody one thing — after I see that gentle bulb go on in a scholar.”
Along with incomes educational honors, together with a 2023 Nationwide Science Basis College Early Profession Growth Award, Wu has additionally been formally celebrated for her educating and mentoring, together with with the Ole Madsen Mentoring Award in 2025.
What does she inform college students confronting extraordinarily difficult issues?
“Be affected person. Begin small. Societal influence is a lifelong endeavor, not one thing to be achieved in a number of years,” Wu says. “It is going to take years to actually perceive what’s occurring and the place the true issues are. Within the meantime, attempt to be useful. Be curious. Ask many questions.”
