Non-Trad PhD to MSFT L63 ML Scientist. AMA.

Microsoft fuckivory
Sep 1 40 Comments

It's almost my one year anniversary in MSFT (joined Oct 2018). I'm giving back to Blind, so AMA.

First, I have a Non-Trad PhD (aka not CS or maths). I made a mistake of not following TC. Wasted 5 years of my life for a useless PhD, and condemned to endless postdoc in the hope of shitty professor position (TC $100k). I have 15 papers from grad sch, 9 first author ones. 3-5x more than the average grad student. My tip for the academia cult: NOBODY GIVES A FUCK about your stupid papers (maybe unless it's a relevant PhD?).

In my 4th year of grad school I woke up, but it was too late to change. I grad in 5th year and did 2 yrs postdoc (TC $46k) because I had no fucking options when looking for a job. Even though I was superstar, I didn't matter. Life outside tech is PURE SHIT.

During this time 'data science' and 'big data' was hot. For the next 2 yrs, I chased it too, but chased deep learning (aka 'AI') as I thought it would be the next big thing. I also stopped publishing in journals. Fuck papers, they weren't getting me what I wanted.

Fast forward to today, I got promoted to L63 and have 140% target bonus. Was hired as L62 at MSFT last year. In 2016, I had 1 onsite and 1 offer. Fuck this! In 2018, I had 10+ onsites and 6 offers! My highest offer was AMZN L5 at $240k. But I took MSFT. My TC is now $220k. Not too bad but could be better. Current YOE is 5yrs PhD, 2yrs postdoc, 1yr at MSFT.

I feel very liberated now that I am outside the ivory tower. I feel fucking loved. For those who want out, I know the feeling. It can be done.

AMA.

comments

Want to comment? LOG IN or SIGN UP
TOP 40 Comments
  • New omgosh
    I’m in a similar situation but with 1/2 the number of publications you had.

    1. How important is LC? How much did u LC?

    2. What kind of knowledge did u have to learn to qualify for the ML role

    3. I took a DS class in college and is trying to refresh myself on it by studying more on DS. Like the usual 10 algorithms of supervised and unsupervised learning, EDA, cleaning dirty data, cross validation. Is this going towards the right direction for ML? Or is this for DS?

    4. How important are side projects?

    5. Was your ML knowledge self studied or did you take a boot camp or classes
    Sep 1 13
    • Apple / Data
      dyh

      Apple Data

      PRE
      Google
      dyhmore
      Can you elaborate the probability and stats questions that you were asked?
      Sep 1
    • Microsoft fuckivory
      OP
      Stats - Explain t-test, what is the central limit theorem. Sometimes it's also a 'case study' that needs A/B testing and you explain how to set it up. Hiccups are thrown and you explain why.

      Probability - Bloody card and dice questions. BAYES THEOREM!!! Lots of tricky conditional probability stuff too.

      Solution is to go brush up on first year stats and probability. I went through some Coursera course and it helped a little.
      Sep 1
    • New / Data JhYt70
      Thanks for the great answers. It really helps me to get better understanding of what to expect in an interview for ML role.

      My plan is to study like a final exam or GRE and apply for ML roles in December (again currently DS). Do you know if there is a better time to apply for these jobs? In general people told me that Q1 is hiring season. Do you have any other opinion?

      Thanks again for advise.
      Sep 1
    • Optum djt1658
      Interesting & congrats. What field was your PhD in?
      Sep 5
    • Microsoft fuckivory
      OP
      Biology.
      Sep 7
  • New / Data JhYt70
    Thanks for sharing your experience, and congratulations! How was your experience at MSFT? What would you suggest if I want to move from DS to ML scientist?
    Sep 1 8
    • Clover Health AsQd52
      How would you define DA, do you consider those with the quantitative analyst/data science position at Google to be doing DA?
      Sep 7
    • Microsoft fuckivory
      OP
      DA is fake data scientist. Mostly BI related. Excel and SQL monkey. Don't know how to code. Uses GUI. Only have bachelor's degree. Thinks linear regression is sophisticated analysis.
      Sep 7
    • Clover Health AsQd52
      I didn't know that about Facebook and Amazon DS, I'm mildly surprised actually and honestly a little skeptical.

      I think often times DS positions are less separated from executives to justify direct business value and impact than ML eng positions. ML eng is also much more SWE focused. So those without a more sophisticated toolset, and who have largely non technical leadership it's easy to fall into the SQL/descriptive statistics only trap. But that's more often in non DS/ML focused start ups ime.

      My connections in big tech are all at Google and Netflix, so definitely surprising to hear about Facebook and Amazon.
      Sep 7
    • Clover Health AsQd52
      I have actually heard actively mediocre things about ds at Apple though.
      Sep 7
    • Microsoft fuckivory
      OP
      That's because the real ML people are under job titles applied scientist (amzn) and ML engineer or research scientist (Facebook).
      Sep 7
  • Nielsen / Eng
    TrilGates

    Nielsen Eng

    PRE
    Google
    TrilGatesmore
    What's your PhD in?
    Sep 1 3
    • Microsoft fuckivory
      OP
      Biology. I used to grow cells too. Obviously I don't put that on my resume anymore (the growing cells part)
      Sep 1
    • Nielsen / Eng
      TrilGates

      Nielsen Eng

      PRE
      Google
      TrilGatesmore
      I mean if you have any comp bio experience that's definitely something to mention. Biology is not easy.
      Sep 1
    • Microsoft fuckivory
      OP
      FYI, I switched to bioinformatics after my first year of grad sch. My shit was just not growing. Lol.
      Sep 1
  • Amazon / Other Valon
    Do you think a graduate degree is truly necessary for a lot of the DS/ML roles out there or is it just a filter to get rid of those who only hold a bachelors?

    Also, would you mind sharing the interview questions you got in 2018 and your prep?
    Sep 1 3
    • Microsoft fuckivory
      OP
      Yes and no. Unless you're researching new algos (and that's only DeepMind, Google Brain, MSR and FAIR), a PhD is not needed.

      The diff between BS and Masters is non existent. If you have BS and can get into FAANG go for that. Masters IMO is letting you roll the dice again. I'm not sure if not having a Masters will be a technical issue like some company requirement for promotion in the long term.

      The PhD does help with solving problems. Lots of time SWEs and new ML engineers ask me how to debug ML stuff when it doesn't 'work'. I don't know to explain but it's easy for me to troublshoot. I suppose after 7yrs+ of working through problems it's some sort of training, but any competent person can learn over the years too.

      In the 10+ companies I had onsite with and probably 20+ phone screens, there is no consistent pattern in their questions unlike leetcode. In some sense, it makes it more difficult because the scope is so big.

      Also there were just so many questions but I would be confident that 70% were a variation of 'explain X' or 'how does X work'. Unfortunately X can be stats, probability, ML, DL (if you apply for DL positions). Coding is leetcode easy to med, some DB/SQL too.
      Sep 1
    • New omgosh
      How many LC easy and how many LC medium did u practice during prep?

      Do you feel it is necessary to also spend time on LC hard?

      How comfortable were you with LC easy and LC medium?
      Sep 1
    • Microsoft fuckivory
      OP
      I did about 100 LC easy and med, and it barely got me through AMZN and MSFT, but got offers at the end so I guess it worked? I didn't apply to G and F cause I knew their LC is tougher.

      Also note that I come from zero LC and CS background. I had to spent almost a month learning data structures, sorting algos, etc.

      I also only spent about 2-3 mths interviewing and preparing and yes that includes learning the basics of leetcode. It was very rushed for me. Don't recommend it.
      Sep 1
  • Peloton AWqC63
    I am in tech but would like to get into bioinformatics. Doing an online masters. I suppose you would tell me to turn back now? :) Do you ever feel you could be more impactful in bioinformatics than at msft?
    Sep 5 5
    • Microsoft .-.-.
      Where are you doing online masters from ? Also how is the pay in bioinformatics?
      Sep 5
    • Microsoft fuckivory
      OP
      Bioinformatics is shit. Pay is shit. WHY would you leave tech?

      This is the most real advise you will get. Anything bio related even biotech and bioinformatics is shit in terms of TC, and career outlook
      Sep 7
    • Peloton AWqC63
      Bummer. I work for a company now where at least it is fitness based so i can feel quite positive about the net effect on society. However if I look back on my career it has been mostly spent on advertising, shopping, or gaming projects. All fairly worthless. I guess my motivation is that bio can be more impactful and I hope to spend more of my day learning amazing truths about life. I understand that is likely pretty naive. Perhaps it is best as a hobby. Also you are doing ML so I imagine that is actually pretty dang cool and rewarding.

      Also I wonder if an impending change to the investment environment would shift the balance toward labor in biotech. Im curious why labor is not treated well in bio. Not like tech wants to pay well, but competition for talent, ability to start your own company with available VC funding, keeps labor in high demand. Things could change 🤷🏻‍♂️
      Sep 7
    • Microsoft fuckivory
      OP
      People in sciences like biology don't chase TC. They have an anti TC culture. They want to feel good and shit like that. If you want to be like that go join them.

      Also profit margins in tech are better and our products are more scalable
      Sep 7
    • Peloton AWqC63
      Word. Dont get me wrong. Super happy for you. Sounds like you found an amazing job and still indeed doing science, with an opportunity to push the boundaries of what humans can do. It makes me sad that our society and values makes you regret spending that time learning the truths of the universe. Certainly also the fault of academic structure as well.
      Sep 7
  • Google / Eng Feeunvr
    What primary languages do you use?
    What primary type of modeling do you do?
    Do you put anything into production?
    Sep 1 1
    • Microsoft fuckivory
      OP
      Python. No R.
      Mostly ML and DL.
      Not before I joined MSFT.
      Sep 1
  • Microsoft
    E8@FB

    Microsoft

    PRE
    Taco Bell, Campbell Soup, Jack in the Box
    E8@FBmore
    Yawn
    Sep 1 0

Salary
Comparison

    Real time salary information from verified employees