How does the program work?
Follow these steps:
- Secure a research opportunity. This typically involves sending cold emails to researchers, which you can read more about below.
- Apply to the scholarship program. After you have secured a research opportunity, and have at least a rough idea of what you will work on, apply by February 17th.
If you are accepted, you will ask you to submit a report in June that explains what you worked on. If you contributed to safety related research and put in the minimum hours, you would receive the $2,000 scholarship at the end of the semester.
In order to be eligible to receive a stipend, you must:
- Be an undergraduate or master's student.
- Secure a mentor. Independent researchers will not be supported.
- Devote at least 10 hours per week to safety related ML research. For a non-comprehensive list of ML Safety research directions, refer to the ML Safety course (the Robustness, Monitoring, Alignment, and Systemic Safety sections) or Unsolved Problems in ML Safety.
- Work on empirical ML safety topics. Conceptual, philosophical, and policy research are outside of the scope of this program.
Step 1: Secure a research opportunity
Determining whether you are ready for empirical ML research
Before reaching out to prospective mentors you should:
- have a solid background in Machine Learning and Deep Learning. If you don’t have this background, we recommend Week 1-6 of MIT 6.036 followed by Lectures 1-13 of the University of Michigan’s EECS498 or Week 1-6 and 11-12 of NYU’s Deep Learning.
- take a topics course in Computer Vision or NLP so that you have exposure to cutting-edge research.
- have a >3.5 GPA. You will be hard-pressed to find a mentor if you don’t have excellent grades.
Going through course.mlsafety.org or participating in Intro to ML Safety could also be helpful (but is not necessary).
Finding a mentor to work with
In order to grow as a researcher, you need to find a mentor. Note that you will very likely help with their research project if this is one of your first research opportunities (make sure it is safety related!). We will evaluate your application according to your merits, but also according to the quality of your mentor. You should select a mentor that has a strong research track record in empirical ML.
Things to note:
- You can typically find people’s emails on their personal websites or their papers.
- Do your research. Make sure you understand what a prospective mentor works on before emailing them.
- Don't spam researchers. Write each email individually and don't send more than 5 per day.
- Don’t send more than 20 emails in total. If no one gets back to you after 20 emails, you probably need to improve your CV.
- You CAN reach out to people at other universities – many collaborate through zoom. This means that if your university does not have good, relevant, or available graduate student mentors, you can still do ML Safety research.
Step 2: Apply to the scholarship program
After you have already secured an opportunity and you at least have a rough idea of what you are going to work on, apply for the scholarship program here. Although we expect most accepted participants to receive the scholarship at the end of the calendar year, this is not guaranteed. You must demonstrate that you followed through with the research you indicated you would contribute to.