data science jobs for physics phd

Diversity & Inclusion
Community Values
Visiting MIT Physics
People Directory
Faculty Awards
History of MIT Physics
Policies and Procedures
Departmental Committees
Academic Programs Team
Finance Team
Meet the Academic Programs Team
Prospective Students
Requirements
Employment Opportunities
Research Opportunities
Graduate Admissions
Doctoral Guidelines
Financial Support
Graduate Student Resources

PhD in Physics, Statistics, and Data Science

MIT LEAPS Program
for Undergraduate Students
for Graduate Students
Mentoring Programs Info for Faculty
Non-degree Programs
Student Awards & Honors
Astrophysics Observation, Instrumentation, and Experiment
Astrophysics Theory
Atomic Physics
Condensed Matter Experiment
Condensed Matter Theory
High Energy and Particle Theory
Nuclear Physics Experiment
Particle Physics Experiment
Quantum Gravity and Field Theory
Quantum Information Science
Strong Interactions and Nuclear Theory
Center for Theoretical Physics
Affiliated Labs & Centers
Program Founder
Competition
Donor Profiles
Patrons of Physics Fellows Society
Giving Opportunties
physics@mit Journal: Fall 2023 Edition
Events Calendar
Physics Colloquia
Search for: Search

Many PhD students in the MIT Physics Department incorporate probability, statistics, computation, and data analysis into their research. These techniques are becoming increasingly important for both experimental and theoretical Physics research, with ever-growing datasets, more sophisticated physics simulations, and the development of cutting-edge machine learning tools. The Interdisciplinary Doctoral Program in Statistics (IDPS) is designed to provide students with the highest level of competency in 21st century statistics, enabling doctoral students across MIT to better integrate computation and data analysis into their PhD thesis research.

Admission to this program is restricted to students currently enrolled in the Physics doctoral program or another participating MIT doctoral program. In addition to satisfying all of the requirements of the Physics PhD, students take one subject each in probability, statistics, computation and statistics, and data analysis, as well as the Doctoral Seminar in Statistics, and they write a dissertation in Physics utilizing statistical methods. Graduates of the program will receive their doctoral degree in the field of “Physics, Statistics, and Data Science.”

Doctoral students in Physics may submit an Interdisciplinary PhD in Statistics Form between the end of their second semester and penultimate semester in their Physics program. The application must include an endorsement from the student’s advisor, an up-to-date CV, current transcript, and a 1-2 page statement of interest in Statistics and Data Science.

The statement of interest can be based on the student’s thesis proposal for the Physics Department, but it must demonstrate that statistical methods will be used in a substantial way in the proposed research. In their statement, applicants are encouraged to explain how specific statistical techniques would be applied in their research. Applicants should further highlight ways that their proposed research might advance the use of statistics and data science, both in their physics subfield and potentially in other disciplines. If the work is part of a larger collaborative effort, the applicant should focus on their personal contributions.

For access to the selection form or for further information, please contact the IDSS Academic Office at [email protected] .

Required Courses

Courses in this list that satisfy the Physics PhD degree requirements can count for both programs. Other similar or more advanced courses can count towards the “Computation & Statistics” and “Data Analysis” requirements, with permission from the program co-chairs. The IDS.190 requirement may be satisfied instead by IDS.955 Practical Experience in Data, Systems, and Society, if that experience exposes the student to a diverse set of topics in statistics and data science. Making this substitution requires permission from the program co-chairs prior to doing the practical experience.

IDS.190 – Doctoral Seminar in Statistics and Data Science ( may be substituted by IDS.955 Practical Experience in Data, Systems and Society )
6.7700[J] Fundamentals of Probability or
18.675 – Theory of Probability
18.655 – Mathematical Statistics or
18.6501 – Fundamentals of Statistics or
IDS.160[J] – Mathematical Statistics: A Non-Asymptotic Approach
6.C01/6.C51 – Modeling with Machine Learning: From Algorithms to Applications or
6.7810 Algorithms for Inference or
6.8610 (6.864) Advanced Natural Language Processing or
6.7900 (6.867) Machine Learning or
6.8710 (6.874) Computational Systems Biology: Deep Learning in the Life Sciences or
9.520[J] – Statistical Learning Theory and Applications or
16.940 – Numerical Methods for Stochastic Modeling and Inference or
18.337 – Numerical Computing and Interactive Software
8.316 – Data Science in Physics or
6.8300 (6.869) Advances in Computer Vision or
8.334 – Statistical Mechanics II or
8.371[J] – Quantum Information Science or
8.591[J] – Systems Biology or
8.592[J] – Statistical Physics in Biology or
8.942 – Cosmology or
9.583 – Functional MRI: Data Acquisition and Analysis or
16.456[J] – Biomedical Signal and Image Processing or
18.367 – Waves and Imaging or
IDS.131[J] – Statistics, Computation, and Applications

Grade Policy

C, D, F, and O grades are unacceptable. Students should not earn more B grades than A grades, reflected by a PhysSDS GPA of ≥ 4.5. Students may be required to retake subjects graded B or lower, although generally one B grade will be tolerated.

Unless approved by the PhysSDS co-chairs, a minimum grade of B+ is required in all 12 unit courses, except IDS.190 (3 units) which requires a P grade.

Though not required, it is strongly encouraged for a member of the MIT Statistics and Data Science Center (SDSC) to serve on a student’s doctoral committee. This could be an SDSC member from the Physics department or from another field relevant to the proposed thesis research.

Thesis Proposal

All students must submit a thesis proposal using the standard Physics format. Dissertation research must involve the utilization of statistical methods in a substantial way.

PhysSDS Committee

Jesse Thaler (co-chair)
Mike Williams (co-chair)
Isaac Chuang
Janet Conrad
William Detmold
Philip Harris
Jacqueline Hewitt
Kiyoshi Masui
Leonid Mirny
Christoph Paus
Phiala Shanahan
Marin Soljačić
Washington Taylor
Max Tegmark

Can I satisfy the requirements with courses taken at Harvard?

Harvard CompSci 181 will count as the equivalent of MIT’s 6.867. For the status of other courses, please contact the program co-chairs.

Can a course count both for the Physics degree requirements and the PhysSDS requirements?

Yes, this is possible, as long as the courses are already on the approved list of requirements. E.g. 8.592 can count as a breadth requirement for a NUPAX student as well as a Data Analysis requirement for the PhysSDS degree.

If I have previous experience in Probability and/or Statistics, can I test out of these requirements?

These courses are required by all of the IDPS degrees. They are meant to ensure that all students obtaining an IDPS degree share the same solid grounding in these fundamentals, and to help build a community of IDPS students across the various disciplines. Only in exceptional cases might it be possible to substitute more advanced courses in these areas.

Can I substitute a similar or more advanced course for the PhysSDS requirements?

Yes, this is possible for the “computation and statistics” and “data analysis” requirements, with permission of program co-chairs. Substitutions for the “probability” and “statistics” requirements will only be granted in exceptional cases.

For Spring 2021, the following course has been approved as a substitution for the “computation and statistics” requirement: 18.408 (Theoretical Foundations for Deep Learning) .

The following course has been approved as a substitution for the “data analysis” requirement: 6.481 (Introduction to Statistical Data Analysis) .

Can I apply for the PhysSDS degree in my last semester at MIT?

No, you must apply no later than your penultimate semester.

What does it mean to use statistical methods in a “substantial way” in one’s thesis?

The ideal case is that one’s thesis advances statistics research independent of the Physics applications. Advancing the use of statistical methods in one’s subfield of Physics would also qualify. Applying well-established statistical methods in one’s thesis could qualify, if the application is central to the Physics result. In all cases, we expect the student to demonstrate mastery of statistics and data science.

Career Profile: Data Science in Industry

The data science in industry career at a glance

Education: MS or PhD in physics or other scientific or computational field or a BS with relevant skills and experience can be sufficient

Additional training: Experience in programming, machine learning, or working with databases

Salary: Starting at $80K - $100K, with mid-career salaries at $160K - $180K

Outlook: The private sector employs over half of physics PhDs and about 95% of those with a bachelors in physics. Specifically, data science is a growing field with many job opportunities for physics degree holders.

What they do

A physicist in a data science job will spend most of their time analyzing data and designing and developing models to predict how something will behave based on data of how it has behaved in the past. Data scientists often work with a team to complete projects. Typical activities include:

Design, develop, and maintain machine learning and other data models
Select, use, and debug existing data models
Perform statistical and data analyses, often to make decisions about products or projected audiences
Conduct research to learn more about the field and to improve model accuracy, including meeting with and interviewing experts
Work in teams to assess project needs and perform tasks

Some data scientists also work on:

Data visualization, e.g. using Jupyter Notebooks/Python
Database management and data quality monitoring
Data infrastructure systems, which consist of all tools and software needed to collect, store, and analyze datasets
Web development
Communication with multiple teams, such as marketing, technical, etc.

Education & background

A bachelors in physics or other scientific/computational field can be sufficient, but a masters or PhD in these fields is often preferred. Programming skills and familiarity with machine learning, databases, and statistics are critical.

Commonly used languages in data science include: Python, R, SQL, SAS, and Scala. Having some knowledge or experience with one or more of these can make candidates more attractive for data science jobs.

Unlike many academic positions, experience in postdoctoral appointments is not considered a prerequisite for data science jobs in most private sector companies.

Additional training

Technical experience in the following can better prepare candidates for data science jobs and increase chances of hire:

Machine learning projects (like participating in a competition like Kaggle.com, an online machine learning resource hosted by Google)
Python, R, or Scala programming language
SQL experience with large databases
Web development projects
Internship in data science or related field

Resources to help build these skills include Coursera courses for gaining knowledge, LeetCode for practicing skills, and Kaggle. The book Cracking the Coding Interview can be a helpful guide when preparing for job interviews.

Effective communication and collaboration skills can also set candidates apart. Students and early career physicists often have translatable skills, such as experience working in scientific collaborations or giving talks at conferences. Additionally, to succeed in industry, one has to be flexible in changing projects and willing to learn new skills. A defining characteristic of jobs in industry is that things move quickly; being able to work efficiently on projects and meet deadlines is key.

When applying for a job in the private sector, understanding the difference between a CV and a resume and being able to write a good resume are very important. For a good tutorial on the difference between CVs and resumes, and for advice on how to write a skills based resume suitable for private sector jobs, please watch our video tutorial .

Career path

Most physicists will start out as a data scientist or analyst, spending a majority of their time writing/developing code. After working for about 5 years or so as an individual contributor, some will move into more senior positions, choosing either leadership roles or continuing to be an individual contributor as a senior data scientist. Other options include being a technical leader or architect, or a manager. In such a role, a data scientist would spend most of their time on project, resource and personnel management. High level management positions in companies carry among the highest salaries for physicists in the private sector.

Deborah Berebichez

Debbie decided that she wanted a life outside of academia and research. She took her smarts to Wall Street and became a quantitative risk analyst.

Jessica Kirkpatrick

Jessica utilizes her astrophysics background through data science to understand the world around her.

Meghan Anzelc, PhD

Meghan Anzelc is an executive leader in data and analytics product development and deployment.

Join your Society

If you embrace scientific discovery, truth and integrity, partnership, inclusion, and lifelong curiosity, this is your professional home.

Solving real-world problems

Cultural challenge, job satisfaction, data science can be an attractive career for physicists.

Open the PDF for in another window
Reprints and Permissions
Cite Icon Cite
Search Site

Toni Feder; Data science can be an attractive career for physicists. Physics Today 1 August 2016; 69 (8): 20–22. https://doi.org/10.1063/PT.3.3261

Download citation file:

Ris (Zotero)
Reference Manager

If different people buy the same items at the grocery store, will their taste in movies also strongly overlap? Can a company recognize when someone tries to make a fraudulent payment? Is a home buyer getting a fair price? Those are the sorts of problems that data scientists tackle.

“Data science is the marriage of statistics and computer science,” says Janet Kamin, chief admissions officer at NYC Data Science Academy. “It is the art of finding patterns and insights in large sets of data that allow you to make better decisions or learn things you couldn’t otherwise learn.” The demand for data scientists is booming across industries—retail, automotive, banking, health care, and more. It’s also growing in the nonprofit and government sectors. (See the plot on page 22 .)

Demand for data scientists is booming. Shown here is the relative growth in US data science job postings. (Data courtesy of Indeed.com.)

In parallel with the data explosion, boot camps on how to deal with the data have mushroomed over the past few years. Most tech boot camps are in coding, Web and mobile app development, and user experience design, but more and more are popping up with a focus on big data and data science. With the aim of helping people transition into the field, the data-focused camps offer various combinations of coding, math and statistics, and machine learning, plus hands-on experience and assistance with landing a job.

“Data in industry has been having a moment for the last 5 to 10 years,” says Berian James, who moved into data science from astrophysics. “It’s transformative, and I felt I would regret not being part of it.”

“There is a huge glut of PhDs who can’t get tenure-track positions,” says Ryan Orban, chief technology officer at Galvanize, which offers data science programs and other tech-sector educational and entrepreneurial services. “We wanted to be a bridge by creating a hands-on, practical curriculum. Industry is looking to hire people who have not just a theoretical understanding but the ability to deliver on day one by building models and solving problems.”

“We give participants three things,” says Kim Nilsson, an astrophysicist and founder of the London-based Science to Data Science (S2DS) boot camp. “We teach them how to commercialize their skills.” For example, boot campers learn how to use Apache’s analytics platform Spark, which is popular in industry and rare in academia. “Second, they learn about working in a commercial environment, with deadlines, intense teamwork, and the culture of business. Academics are meticulous, they want to get things 110% right. In business, 80% is good enough. And third, we help them form a network.” (See the interview with Nilsson on Physics Today ’s website.)

Boot camps typically last from 5 to 12 weeks. They can be on site or online. Some cater to people with a degree in a science, technology, engineering, or math (STEM) field. Some are extremely selective, accepting just a few percent of applicants, while others are open to anyone who can pay; the cost varies from free to around $20 000.

For example, the five-week-long S2DS camp, which accepts mainly STEM PhDs, costs £800 ($1000), including housing. Companies pay a fee to have S2DS participants work on a project; in return, the companies get the results from the project and the opportunity to recruit the participants. NYC Data Science Academy charges its broader mix of participants $16 000 for a 12-week course. “We look for people with a STEM background,” Kamin says. “But we will take a risk on people who are uniquely talented and motivated—like lawyers and people in marketing and psychology.” The breakdown of boot campers tends to be about one-third PhDs, one-half masters, and some bachelors, she adds. In a third model, di-Academy in Brussels trains people sponsored by their current or prospective employers, who pay about €15 000 ($16 700) per person for a 12-week data science boot camp.

The typical boot camp attendee is in their mid 20s to early 30s, but the age range is from the early 20s up into the 50s; women make up from 20% to nearly 40% of attendees. Most participants need to know at least some programming going in.

Claire Lackner , a former astrophysicist, at Insight Data Science, where she got the training and connections she needed to move into a data science career. She is now a data scientist at Element Analytics.

Attendees and instructors at the close of a 12-week boot camp at Data Science Retreat in Berlin. The attendees had just presented their individual projects to recruiting companies. Founder Jose Quesada is second from right.

The average annual salary for a data scientist in the US is $117 000, according to the job-tracking website Indeed.com. Testimonials on boot camp websites showcase physicists turned data scientists working at startups and at well-known companies like Twitter, LinkedIn, and Google.

Kamin notes that physicists rarely come in knowing what they want to do after they complete boot camp. “They are not the ones who say, ‘I want to analyze medical records.’ Instead, they come in saying, ‘I love breaking down big problems into small pieces.’ They are agnostic about what they do; they just want to solve problems.”

Benjamin Arar earned his bachelor’s degree in physics from Princeton University in 2013. A year later he was frustrated: He was getting into machine learning on his own, but he was getting nowhere in his search for employment. Early this year Arar attended the inaugural 12-week session of the RSquare Edge boot camp in New York City; the startup gave him a partial scholarship for the $19 000 program. “It was intense,” he says. “Every four weeks we switched to a new set of classes with homework and a project.”

Within two months of completing the program, Arar had two job offers: one from a startup where he would create predictive models for media companies, and one from a company where he would build real-time analytic and machine-learning tools for marketing applications.

James, the former astrophysicist, first heard about the opportunities in data science when he was a postdoc at the University of California, Berkeley. He applied to Insight Data Science for a free seven-week program for PhD scientists. After attending the boot camp in Palo Alto, he got a job at the San Francisco–based payment-processing company Square, where he wrote algorithms that used machine learning to detect fraud. Four years on, he now manages a team of about 20 data scientists at the company.

James was originally drawn to physics by his desire to learn and understand the universe. He was happy to find that the opportunities for learning in industry “accelerated,” he says. “It’s not as much depth as in an academic career, but the variety of things to work on is big—fraud is just one component; marketing, sales, these will be rich fields for decades for data science.”

Martina Pugliese began looking around at career options as a doctoral student in physics at Sapienza University of Rome. Her PhD work applying mathematical theories from complex systems to linguistics “was the embryo of my path in data science,” she says. “I did a good amount of data mining and numerical modeling, but not much machine learning.”

Attending an S2DS boot camp in 2014 helped in two ways, says Pugliese. It provided access to companies, and by doing a real-life project during the boot camp she proved herself as a data scientist. She is now at Mallzee, a startup company in Edinburgh, Scotland, where she works on an app that personalizes clothing recommendations and sells the items it lists.

Before the recommendations can be personalized, though, clothing data from retailers’ websites need to be standardized and classified. For example, grabbed text might read, “Beautiful pencil skirt in blue, wear it with ankle boots in brown.” Pugliese uses the probability of words appearing together and applies machine learning to train an algorithm to recognize the product as a skirt. In addition, she feeds data from individuals’ preferences—collected from their viewing and buying histories—to create an algorithm that will display personalized recommendations.

Although Pugliese has never been interested in clothes, she is interested in understanding people’s behavior. Her work, she says, “uses an enormous amount of data, and you can see the psychology of people. And you never get bored. The problems are not meant to be solved in months or years. They are meant to be solved at a quick pace. It gets messy, but it’s fun.”

Data science, says Pugliese, is so fashionable at the moment that it presents a cultural challenge: Data scientists have to get up to speed in the field they are working in to be able to ask the right questions, and people on the business side have to learn what data scientists can actually glean from the data—they should not expect a “silver bullet.”

Luigi Scorzato is a data scientist at the consulting firm Accenture in its Geneva office, where he works with clients to improve their computing architectures and their models for handling growing amounts of data. The work, he says, “is not less intellectually challenging or interesting than what I was doing in particle physics.” Most of the time in particle physics, he says, “you are involved with problems that are not really fundamental.” Now, he says, “what I find most interesting is to see problems solved and to help in innovation.”

The technologies used in industry change fast, as do the problems. But, says Scorzato, research, in particular research in physics, will remain extremely useful for implementing the most innovative solutions.

Citing articles via

Online ISSN 1945-0699
Print ISSN 0031-9228
For Researchers
For Librarians
For Advertisers
Our Publishing Partners
Physics Today
Conference Proceedings
Special Topics

pubs.aip.org

Connect with AIP Publishing

This feature is available to subscribers only.

--> Twitter -->