Bright Futures Q&A: Debbie Berebichez

28. September 2016

OPN recently had the opportunity to talk with Debbie Berebichez about an increasingly hot topic for physics graduates: careers in data science.

Berebichez received her Ph.D. in theoretical physics from Stanford University, Calif., USA, but opted to not pursue an academic career. Instead, she sought out roles unexplored by most scientists. Berebichez first combined her love for communicating science in her online video series, “Science Babe: The Science of Everyday Life.” In these videos, she explained basic science, like the inner workings of a microwave oven. The videos attracted the attention of Oprah Winfrey, and in 2007, Berebichez was invited to be the keynote speaker at a conference on women’s leadership organized by Winfrey and her team. From there, Berebichez went on to host scientific television shows—even while working a day job on Wall Street.

Berebichez is now the chief data scientist at Metis, a data science-training company. Here, she tells us of her unique path to this burgeoning field and how other physicists can make the same transition.

OPN: Many would assume physics and data science are quite different. Tell us about your journey from one field to the other.

I decided to leave academia in 2009, and that’s when I became aware of “quants”—physicists and mathematicians on Wall Street.

I met physicists who were happy applying their quantitative skills on Wall Street, so I decided to try it. I first worked for a year for a quantitative hedge fund. I enjoyed it. What I was doing there—even though it didn’t have the name of data science—was essentially data science … I found the math fascinating and challenging.

OPN: You then spent six years on Wall Street. Can you elaborate on how you were eventually introduced to data science as a field?

I heard of data science late in the game; I guess I had never heard the term. My friend Hilary Mason, a renowned data scientist, invited me to speak at a conference called DataGotham. I talked about my work in finance. People approached me at the end claiming that what I was talking about was data science. That was funny, because for me it’s always been quantitative science—I didn’t really know what data science was. I started to find out more and more.

I left Wall Street, and it was quite easy, actually, to find a job in data science. They really crave people with physics backgrounds. Plus, if you’ve had some experience with Wall Street, they really like that combination.

OPN: Metis, your current company data offers science boot camps—intensive courses lasting a short time. These seem to be quite popular. Can you tell us more about them?

Our boot camp is a 12-week immersive program held in either New York, N.Y., USA or San Francisco, Calif., USA. Our instructors are experienced senior data scientists. They teach students about Python (a programming language), statistics and algorithms. Students also learn about machine-learning, deep-learning and big-data tools such as Hadoop and Spark.

The boot camp is structured around project-based learning. Students complete five projects throughout the 12 weeks. At the end of the program, students present their final project in front of an audience of companies that will hopefully hire them.

OPN: How does one become a student at a (Metis) boot camp?

We have an admissions process that’s just like a university’s. Applicants get two interviews with instructors where they (the applicants) answer technical questions. We admit about 35 to 40 percent of our applicants.

If we feel that an applicant is going to struggle in a boot camp, we’ll reject them. We let them know where their weaknesses are so that they can apply in the future. We’ve had many people who’ve come back to us that way. We also give about 60 hours of pre-work, to get participants up to speed with what they’ll be learning, and to homogenize backgrounds.

OPN: Boot camp sounds intense. How do students adjust?

We deal with the “imposter complex” quite a bit the first few weeks. We tell students that we want the water level to be at their neck, so they’re not completely drowning, but it’s not the shallow end where people can comfortably walk. This way everybody feels challenged.

Boot camp feels like an incomplete process, because it doesn’t feel like you master everything. But that’s part of what data science is, since it’s such a complex field. You’re never going to know everything; you’re never going to master all the algorithms. As long as students are comfortable looking things up and thinking on their own, then we’ve done our job. But it is a challenge.

OPN: If somebody’s coming into data science from a physics background, what are the holes that somebody in that position might need to fill?

You will realize—if you try to move into data science—that physics is an immense gift. Physics is the basis for so many things; it helps people acquire the essential skills of data science—how to solve problems and how to communicate the solution of those problems to stakeholders—no matter what field you’re in. I haven’t seen any other type of preparation that is better at taking the plunge and solving problems than a physics background.

OPN: What is the best way to prepare for a career in data science?

Boot camp is certainly a great option; we’ve had many physics students and other quantitative backgrounds come through the boot camp run. The biggest challenge is the softer skills like communication … it’s almost like they get rusty with that after spending many years in the lab or in academia.

Besides the boot camp, there are plenty of resources out there. There are many Coursera courses online—even universities are getting on the wagon and offering data science courses (though my own view is that those tend to be very slow compared to the boot camps).

OPN: Is there a place in data science for those who are further along in their careers? Say a mid-career professional in optics wanted to make a career change—would data science be accommodating to someone in that situation?

I would venture to say that Metis’ oldest student was close to 60. The incredible thing is that the salaries that you start with in data science are a lot higher than the ones that people have in academia or research.

It’s very interesting to see people who are mid-management or at the executive level in their careers come and be humbled by this boot camp—it’s challenging ... They definitely go through this sort of shaky month or two. They question if they’ll ever come out successful on the other side.

OPN: But you’ve had mid-level or executive level professionals find success in boot camp, yes?

Yes. Many of the students, once they get the hang of it and learn it’s okay to get your hands dirty, and not be the best for three months—and really try to learn as much as you can—are the students that are older and get amazing positions. They renew themselves and have this new lease on their professional life, because they never thought at 45, 50 or 55 they would be able to get jobs.

OPN: What are the big growth areas in data science going to be?

There are many cutting-edge techniques. For example, natural-language processing is applied to many different fields­­—from marketing to artificial intelligence (AI) and spatial analysis. Medical device companies use data science. Anything that has to do with what’s called the “Internet of Things,” like putting sensors everywhere to optimize climate control in a factory, or to optimize the driving of a driverless car, also use data science.

All the independent AI stuff requires an immense amount of data science power … it’s a huge, booming industry hiring data scientists. Therefore, anything that’s related to products that collect data—and needs to be analyzed for insight—requires data science.

OPN: Data science has also made its way into traditionally less technical fields. Tell us about some of the advances there.

There’s quantitative marketing. Many online ad agencies struggle to know the impact of their ad campaign; they want to have smart advertising. They’re not happy with simply putting an ad on TV and not knowing. Obviously, that industry has evolved a tremendous amount.

Even trading companies like DataMiner (who trade based on social network data) do a sensitivity analysis to see, for example, what items are becoming popular on Twitter before Black Friday. They apply data analysis techniques to know how to influence the market and sell different products.

OPN: What sort of companies are hiring data scientists?

All the big online companies, like Facebook and Google, are constantly looking for data scientists because they want to be able to recommend products to people, and they want the recommendations to be refined and targeted. They use machine-learning methods, like collaborative filtering and classification algorithms to find people like you. They’re recommending what they have bought or find your historical pattern and recommend products based on that.

OPN: What sort of innovative work is coming out of data science?

I find deep learning to be fascinating. It’s part of this cutting-edge area that combines machine learning with neural networks. Neural networks were something that physicists tried to use many years ago, but didn’t have the computing power. Now it’s a way of finding the answer to deep questions, including science questions. I believe IBM’s Watson may be using this with their successful bio part. That’s an area that’s more cutting-edge, and companies that are doing virtual reality and AI are getting into it.

OPN: You’re speaking at the Strata conference on how data can be misleading through the poor use of statistics. Can you tell us more about this?

I think that’s a question that is deep in my heart. Coming from physics, I want to teach data science not as a memorization book, but as a critical thinking exercise. But that experience of people working very closely with data—and being incredibly savvy at manipulating data—and yet not knowing what they’re doing, that is everywhere in data science. It’s really alerted me to the necessity of having people think through their datasets and what they’re doing, before they become savvy at data mining.

Dr. Debbie Berebichez is the chief data scientist at Metis. She is also a physicist, TV host and STEM advocate. She graduated from Stanford University, Calif., USA, with a Ph.D. in Physics, and received undergraduate degrees from Brandeis University, Mass., USA. Berebichez is a co-host on Discovery Channel’s Outrageous Acts of Science TV show.

Photo by Bruce F. Press Photography

Career Path, Nontraditional Science Careers, Profiles

Comments are closed