My internship at Intel Labs Munich: 14 take-away lessons

Week ago I have finished my 6-month journey at Intel Labs Munich. First, couple of words about what is it.

Last years become quite popular among AI-related corporations (Google, Microsoft, Intel, Facebook, etc.) to host internships for PhD students. It is quite different from regular internship. The main idea is to do a scientific paper (CVPR-level) on the topic given by company, but somehow related to your PhD topic.

It is good for student, because:

  • safe demo-version, of what work in corporative R&D looks like. If you both like each other, usually you will got an job offer much easier than default screening process.
  • expands your experience and CV
  • new connections
  • well, resulting paper itself.

It is good for company, because:

  • safe way to recruit junior researchers (if they are good) in competitive market
  • paper
  • PR

So I decided to go for it and, just in time, got proposition from Vladlen Koltun to try myself at Intel Labs. Group is pretty small, but produces very high-level reasearch. OK, lets finish introduction and go stright to the point(s).

  1. Vladlen lab standards are pretty high and that is about everything: work, experiments, text, presentations. Our submission is more polished, that any of my previous camera-ready. That is cool and tough. I definitely would recommend you to go for internship here.
  2. Lab has methodology of writing papers and lots of LaTeX/matplotlib snippets. That helps to make good-looking paper at affordable time costs.
  3. Everyone presents her/his research on lab meetings at least once — and as if it was real external presentation. Goal: to get feedback and internal reviews. And to make everyone in group aware about your results. Moreover, intermediate deadline (presentation) quite helps to shape work one way or another. I like that.
  4. Don`t try to advance state-of-the-art in new field by new (for you) method during internship. I tried this and failed. Not miserably — we still produced good work, but not as great as I hoped for. You can try new area alone, or new methods in known area, but not both — don`t repeat my mistakes.
  5. Dig deep and straight to the core. Question everything. My favorite example of such attitude is “Learning to see in the dark” CVPR 2018 paper. Task: get good quality photos in very low light. While other researchers take RGB images as granted, Chen feed into CNN not even RAW image format, but something even less pre-processed. This allowed to obtain quality far beyond previous works.
  1. 6 month is never 6 month of work time. You will have corporate trainings, onboarding, conferences (I visited both CVPR and ECCV — so minus 3 weeks in total), other interruptions, etc. Better to count on 4–4.5 month.
  2. Hour spent on writing good and clean code saves you a day in deadline time. And lots of nerves.
  3. Think, plan and code in the morning. Read, do meetings, debug in evenings.
  4. Macbook sucks. Never got used to it, especially keyboard shortcuts and lack of “delete” key.
  1. Intel helps with all needed bureaucracy. That is great, because Germany is…well, lets say, demanding in that sense. Still, it is not possible to go for 6 month with family: reuinion visas are only for >=1 year contracts. Lets face it: the most of academia lifestyle, including internships, are not designed for people with family. And not only academia. I hope, that will change somehow, but now it is as it is.
  2. Munich is an expensive city. Rent rate is the highest in Germany, moreover it is still hard to find a flat: there are more people than places, so landlords have freedom of choice. Actually, not that hard to find something for you alone: flat for 1 or even room in shared flat. But for family — Munich landlords think that for couple + infant you need 3 bed apartment. Add here 6 month period: too big for airbnb, but too small for typical rent.
  3. Coffee sucks here. If you love nice, acidic espresso…brace yourself.
  4. In Ukraine there are lots of 24/7 shops. In Prague, there is at least one Vietnamese food shop nearby, which works until late. In Munich, 8pm is deadline. And if you got used to stereotypical IT-person schedule: come to work at 11, leave at 8…I have very bad news for you.
  5. Munich technical museum is great. “English Garden” park is great.
  6. Best pizza in the city — Lo Studente.

So, that is all. Would I recommend you to do the same? If you are single - yes for sure. If not - well, it depends on lots of things: family, citizenship(s), etc.

If you have questions — write them in the comments. If it is not under NDA, I will answer.

Finally, thanks to all my colleagues there: Alexey, Vladlen, Adel, Rene, Katrin, Ozan, Gernot, Stephan, Henry.

Lab dinner in July 2018. Taken from Vladlen website

Computer Vision researcher and consultant. Co-founder of Ukrainian Research group “Szkocka”.