Misc.

Resources for Breaking Into ML Research

List of opportunities and advice I found either online or through talking to researchers. I’m personally interested in industry research, so you might not find resources or advice geared towards an academic career. I will try my best to keep on adding to this list.

Opportunities

  1. AI Residency programs: many companies had these but most stopped in 2023. Watch out for them incase they or other similar roles open in the future. Concurrently, Google has opened many student researcher positions which, unfortunately, are exclusively for students (at all levels i.e. BSc, MSc, PhD).

  2. Cohere Research Scholar program and Cohere open-science community. I highly recommend the open-science community, it’s how I hope the future of research looks like.

  3. YAAIR Fellowship (For African MSc/PhD grads).

  4. Allen Institute Pre-doctoral fellowships. Do look at the institutions past fellows came from. The institution names were mentioned along with their fellows’ names in the previous version of the website (which suggests an emphasis on the ‘fanciness’ of your school).

  5. ML Collective Community

  6. Open Office Hours. List of Researchers with Open Office Hours you can talk to about research, careers, grad school etc.

  7. Collection of SOPs (Statement Of Purpose) written for grad school applications.

  8. Black researchers community

  9. Women in Machine Learning community

  10. Fatima Fellowship. Work remotely with PhD students or PostDocs in top AI research labs. If you don’t have previous research experience but want to apply to grad school, this is your chance.

  11. Masters programs for Africans in Africa: AIMMS, CMU Africa

  12. For those interested in mechanistic interpretability, there are several opportunities and resources like:

    1. MATS program
    2. ERA-KASL Safety Internship
    3. ERA Fellowship
    4. ARENA tutorials
    5. 80,000 hours


    There are many more resources and companies working on mechanistic interpretability and AI Safety. To find all of them, look for programs or companies Open Philanthropy funds. You might observe all of them have one common denominator: they’re part of the Effective Altruism movement (train your ears to hear about AI X-risk regularly).

Advice

  1. Advice I got for landing research roles without a grad school degree (from Research Scientists): It’s very hard for reasons beyond technical qualifications. What you can do is target engineering roles then transfer internally. The semi word for word advice was: ‘I don’t know why enough people don’t do this, but you can transfer to research roles from engineering roles in any company if you show interest. But keep in mind that the line between research and engineering is very blurry and only getting blurrier’. Another researcher: ‘You can also target smaller companies who do not put much emphasis on degrees and titles….also there are many PhD grads working as Research/Software Engineers in research teams (understand your competition).’

  2. ‘You gotta work on the right problem at the right time in the right way, otherwise you’re doomed.’ ~ Richard Hamming

    I surveyed the linkedin profiles of hundreds of researchers to understand which ones become audibly successful and which ones do not. It’s not only about how smart you are, you have to choose a timely problem to work on. My observation: most researchers who graduated with a PhD pre-deep learning explosion or those who didn’t work on hot problems ended up as engineers (either product focused or research focused) while those who worked on deep learning continued doing/leading research projects in research-focused labs (which are the ones you probably hear about).

  3. Some general advice on navigating research life. There are many researchers who put their experience out on the Internet. These are very valuable, as they dispel many unrealistic expectations about what grad school and research are like. A quick perplexity search provides lots of links.

  4. Advice on cold-emailing.

  5. Your work/research is not going to publicize itself. Read ‘The Worlds I See’ book by Fei Fei Li where she talks about the struggles she went through to garner attention to ImageNet (No, people were not begging for access to the dataset).

  6. Having a PhD makes your opinions more valuable to the public and eventually to yourself. You start your sentences with ‘it is’ rather than ‘I think’ (based on my observation). Without it, anything you say or write will be heavily scrutinized (mostly ignored). Be ready to deal with that.

  7. ‘Are you happy?’ I have been obnoxiously asking this question to most researchers that I talked with in the past year (I clarify that I’m asking career-wise). I don’t know why I keep asking this question as I don’t expect them to tell me the truth anyway. But I like to see them reflect on their lives in a few seconds and in someway I get my answer through listening beyond the words. On a side note, I personally think that happiness is a choice you make rather than a reaction to your circumstances.

  8. ‘First make decisions and then make them right.’ ~ Ratan Tata

Resources for ML interview prep

From the info I gathered, Research Scientist interviews are much easier on the coding side, so this is probably an overkill. For ML/Research Engineers however, you have to be able to solve most of these questions within a limited time frame while talking through your thought process to an engineer who is there solely to expose how crappy of an engineer you are (jk😶).

  1. ML Practice problems. This one is a must. I can easily pick a handful of problems from this list that I was asked in interviews for research focused roles.

  2. Big tech companies still ask LeetCode style questions for interviews, but there are many smaller AI companies that don’t, and focus on 1 more heavily instead. NeetCode roadmap is very useful to structure your practice.

  3. Some companies conduct pre-screens on CodeSignal.

  4. ML Interview Guide

  5. Want to learn how to write cleaner ML code? Here is a repo for that.