As a Research Scientist at Lyft, Timothy Brathwaite spends his time building algorithms and using machine learning (ML) techniques to support the company’s transportation app. Born in Brooklyn into a Panamanian family, Timothy spent a lot of his youth cycling around the streets of NYC and solving math problems. His love for cycling and math eventually led him into the field of Urban Planning and Transportation, where he recently graduated with a PhD in Transportation Engineering at UC Berkeley. His dissertation and research centered around using math to encourage more individuals to bike.
As someone who is early on in their career, Timothy is excited to use ML techniques to help shape a positive future in the field. Read on to learn more about his role at Lyft, how he started using ML techniques in his work, and who his role models are.
We interviewed Timothy as part of AI4ALL’s Role Model in AI series, where we feature the perspectives of people working in AI in a variety of ways.
As told to Eunice Poon of AI4ALL by Timothy Brathwaite
EP: As a Research Scientist at Lyft, what is your role and what are your responsibilities overall? What does a typical week look like for you?
At Lyft, research scientists are dispersed amongst product teams. I work mainly on the locations team and I do a small amount of work on the bike and scooters team as well. Overall, my main projects have revolved around using ML techniques to suggest good pickup locations for individuals and to predict the number of scooters or bikes that we expect to be ordered at various locations throughout the course of a day.
For me, a typical week gets broken down as follows. About a third of the week will be spent on miscellaneous tasks; conducting interviews/reviewing candidate homework, 1-on-1’s with team members, team meetings, etc. Then around half the week is spent working on current projects for the financial quarter, and this is where I work with data and build ML models. Then the rest of the hours are spent checking on project feasibility and estimated impact for the projects coming up, and pie in the sky ideas of what I think may be beneficial to the company.
EP: How did you decide to pursue degrees in Urban Planning, Mathematics and Transportation Engineering?
When I was in high school, I got into a collision with a motor vehicle. I was knocked off my bicycle because a parked car opened its door onto the street and essentially I was “doored” straight into traffic. After that incident, I went home and wanted to look into who makes biking better in cities, and upon searching this in Google, found urban planning as the top search result; so I said “great, I’ll do it!”.
Studying math happened because I really enjoyed it all throughout grade school. Additionally, in freshman year of college for some reason I decided not to take any math courses, and I hated it! So after that first year, I decided to complete a minor in math and also a minor in anthropology, just because it interested me as well.
The decision to do transportation engineering in graduate school came from wanting to combine a love of math with trying to get people to bike more. When I found out about “mode choice modeling” I was obsessed with the idea of creating some cool mathematical model of how people choose how to get around, and then using that model to guide how one planned a city to try and maximize the amount of people biking.
EP: Can you tell us how you learned about ML, and what made you interested in pursuing a career in the field?
I learned about ML at the end of my first year of graduate school. I had a bunch of electives to fulfill, so I was actively searching for courses I could take that would give me the largest “tool belt” possible. I didn’t necessarily think that the methods I learned in transportation engineering, such as discrete choice modeling, were the be-all and end-all, so while I was casting a wide net, an older student in my research group — Dr. Akshay Vij — recommended the class “CS 289A: Introduction to ML”.
That class ended up being the single course that taught me the most during my time at Berkeley. It also piqued my interest in applying ML to problems in transportation because it showed me how incredibly useful ML could be.
EP: What are some of the important things people should be doing to create a positive future for AI?
When I think of some important things people should be doing to create a positive future for AI, three things come to mind. First,
I think we need to have a very frank and open discussion about values, and I mean that in not just a broad societal sense, but also in a very technical sense.
For instance, at the risk of overgeneralizing, I feel like any project that involves ML/AI should begin with open and quantitative discussions about how typical accuracy or business metrics are valued against “diversity” or “fairness” metrics. But there’s never really an open discussion about our values and trade offs in this way and I think that needs to happen before we can make any meaningful progress to having a future for AI that is more fair and more useful to large groups of people. There’s still work to be done on this point, but so far, one of the perks of being at Lyft is that the company culture encourages employees to bring up these and any other tough questions that they think could make the product even better.
The second thing is that even after you’ve had a discussion where you can value and weigh one type of metric against another, there needs to be a way to estimate these ML models to account for the fairness criteria. The unfortunate reality is few will attempt to build “fairer” ML models if they do not have well-implemented/easy-to-use technical tools that facilitate this effort.
The last thing is, even if you have a nice talk about all the trade offs you’re willing to make, and you have ways to impose these constraints and trade offs on the models you build, you need people who care about this in the first place,
and that’s where I think AI4ALL’s mission is critical in trying to get a more representative workforce in AI, and that will hopefully lead to more people in the field, and a wider range of ideas on fairness and diversity.
EP: Who were your role models growing up?
Between elementary and high school, I don’t think I had role models as one would traditionally use such a term. In general, during those years, I just wanted to be really good at something (or lots of things). I most strongly recall the existence of role models once I got to undergrad and graduate school. Even then, my role models were rarely ever role models in an overall sense. Most of the time, I just wanted to emulate certain aspects of people’s lives when I thought they were great examples of the qualities I wanted to see in myself. I figured I could extract the good and ignore the bad.
When I think of “role models” in this sense, there are quite a few people that come to mind. For instance, Janette Sadik Khan, in how she was able to maneuver through the bureaucratic labyrinth that is the public sector in a major city to bring about the most remarkable transformation in on-street bicycle infrastructure in New York City. Kenneth Train and Michael Jordan (the statistician/computer scientist) were huge role models in how they were able to explain complex ideas in an approachable way. Despite the fact that he was a reported racist that wouldn’t teach black people, R.L. Moore was a role model for the magnitude of educational impact I would love to have and for his dedication to and skill in growing researchers who could think creatively and independently.
Other names come to mind, such as Susan Athey and Carlos Daganzo who constantly reinvented themselves long past tenure, Trevor Hastie and Robert Tibshirani in the way they contributed so much to statistical methods and the software R, Cal Newport who acted like a craftsman in how he approached his work, Andrew Gelman in being a Bayesian even when it’s not en vogue in his academic department, Yoshua Bengio in how he continues to make great use of “simple toy problems”, and Richard Hamming in how he continually socializes with others to find the “big problems”.
EP: What advice do you have for young people who are interested in AI who might just be starting their career or academic journeys?
First, I’d recommend that you never get too far away from math! Take courses in school that will require you to work out the math of all the methods you use. After school, take online courses and work through textbooks on your own, trying to recreate ALL the math and code. This will allow you to really understand what one’s methods are doing, the assumptions they entail, how they may fail, and when other people are using these methods inappropriately.
Second, read research articles widely and continuously. This will help you avoid becoming stagnant and will help you grow technically at a fast pace. It’s almost comical to think that one knows everything about every method, so reading is one way to stay abreast of issues with and improvements to the methods one uses.
Third, try to be useful to the quantitative public (people who use and research statistics and ML methods). We all make use of ML methods to solve problems in our work and study. Along the way, we inevitably face situations where the current state of the art and practice falls short of our needs. Instead of just giving up and saying “I can’t do what I wanted to do”, we would all be better off as a community if we each made a habit of digging in and improving the methods we use such that they actually work for us. This may not always be successful, but our communities would be better off if we all tried to improve the existing methods where possible AND if we also shared these improvements via open source software with written descriptions/papers of what we’ve done.
About Timothy
Timothy Brathwaite is a research scientist at Lyft, Inc. Advised by Professor Joan L. Walker, he earned a doctorate of philosophy (PhD) in transportation engineering from the Civil and Environmental Engineering department from the University of California (UC) at Berkeley. Aiming to predict the demand for bicycling under various policy scenarios, Timothy’s research made methodological improvements to discrete choice models to account for omitted roadway variables, traveler “irrationality,” and the typically low number of cyclists in household travel surveys. He was the Outstanding Graduate Student of 2017 for the University of California Center on Economic Competitiveness in Transportation and a 2016 UC Berkeley Outstanding Graduate Student Instructor.
Previously, he received his Master of City Planning and Master of Science in Civil Engineering from UC Berkeley and his Bachelor of Science in Urban Studies and Planning from the University of New Orleans. Before Lyft, Timothy worked with transportation consulting firms (Fehr and Peers, and Cambridge Systematics), with the bicycle facilities program at the City of Oakland, and with the nonprofit “Bike Easy” in New Orleans.