Props to the author--landing your first deep learning position is very difficult. There are many many engineers who want to do deep learning but don't know how, and a few employers that badly need deep learning experts. The employers are (understandably) skittish about taking a chance on hiring someone without expertise and assuming they can train up. If you can show up with proof that you can train performant models, then you'll have a much easier time. Certainly the author put in his time to build up this proof.
By-the-by, if you want to work on autonomous vehicles and aren't too attached to being neck deep in the deep learning parts, nearly all autonomous vehicles companies need an army of engineers for data handling, and on-vehicle performance.
(Vladimir, if you're reading this, I'd be curious to hear what your $$$ investment was for your kaggle competitions).
I always find it funny when I see some British recruiters' posts in my LinkedIn network regarding "awesome Deep Learning opportunities" who require a PhD from a top school, publications at top journals/conferences and offer 65k GBP. Or some "super hot" Deep Learning startups seeking engineers for $3500-5500/month before taxes. For some, having Stanford Deep Learning on resume means one would satisfy minimal requirements for fixing bugs in their crappy robotics system (and they are nothing like Tesla).
It’s a class thing in the UK. I recommend anyone who is any good to jump on a plane to the Bay Area. Same goes for Australia. They refuse to pay competitively and then wonder why they can’t find good talent.
Here in the Netherlands they will pay competitively... for management positions. Even middle management gets compensated very well. Engineers not so much.
I am also a dev in The Netherlands, while a lot of data is available for devs in the US I can't find something similar for NL. What would you consider a competitive pay for devs in NL? (Assuming we are talking about the Amsterdam area for the junior, medior resp. senior levels)?
Seems like an opportunity? Start your own firm, pay a proper salary to good engineers, attract the best local talent in droves and become massively competitive.
On the other hand, I frequently get approached about Machine Learning positions that I'm utterly unqualified for (I did do a maths undergrad degree, but I've never done any machine learning whatsoever).
By simply compensating engineers more, do you think it would be possible for Britain to have a bigger tech scene? In other words, as an American company, would it be beneficial to post jobs in Britain that are above market rate but still less than the US? As many companies are backtracking from India this could be an opportunity
Some US companies partially do this in Ireland, I've heard anecdotally, where salaries are also much below the US. It's a much smaller market obviously, but it does happen.
I’ve seen this done - with UK and US employees working together on the same remote team for an SF-based startup, but it resulted in pressure to pay US employees below market rate.
The most important factor is that northwestern Europe is the least corrupt area in the world. Salary normally is inversely proportional to corruption levels, USA is an exception.
I love posts like these. The author really has awesome persistence and great sense of initiative. Its really good that Kaggle now has GPU training otherwise AWS or GCP will just burn a big hole in your pocket. Handson experience with this some time back.
The credentialism in deep learning is absurd. This guy has a PhD in theoretical physics from UC Davis, he's winning competitions on Kaggle, and he can't even get an interview.
I remember that moment to this day. I am standing at the scene. The organizer is preparing a check and some gifts. Alexey and I emerge victoriously, but I can’t help but feel frustrated by the apparent absurdity. How did this happen that some random dude that does not have domain knowledge in medical imaging got first places in both challenges? And yet, people that make money for a living working on medical imaging use much weaker models?
I asked the audience: “Do you know where do I work?” Noone, except one organizer who checked my LinkedIn profile knew. I told them that I work at TrueAccord, a debt collection agency and that I did not train deep learning models at work. I lamented that I was not able to break away from this model because HRs in Google Brain and Deepmind did not even look at my resume.
I don't know where you got the idea he couldn't get an interview.
>During that same period, I had failed an onsite interview at Descartes Labs, and a few technical screens.
>Somewhere around that time, I was invited for the onsite interview with NVidia, which I also did not pass. One of the issues that I had was my limited knowledge of how 2D object detectors work.
>The next company on my list was Tesla. The recruiter contacted me because of my Kaggle achievements, which did not happen often. I passed take-home tests, tech screenings, and an onsite interview. The next steps were the background check and approval of my application by Elon Musk. I did not pass. [He'd violated a pretty bullshit sounding NDA]
>The company that was organizing the competition was called Planet Labs. They had an open DL Engineer position, I asked about it and was invited to an onsite interview. I failed again. The feedback — not in-depth DL knowledge.
He got a number of interviews, which he didn't pass. Is it really so surprising that Google Brain and Deepmind are going to be pickier than most?
I guess my question is how did he win the competition if he doesn't have enough knowledge to pass the interview? Maybe the interviews really aren't great at actually discerning who can create great DL models? I mean it's possible there are other reasons as well - but the guy objectively beat a bunch of credentialed people in an open competition.
There are a bunch of options here. Maybe they were looking for someone to do novel research. Maybe they wanted someone who can communicate with business partners. Maybe the competition is just people in the middle of their degrees spending an evening on it. Maybe it is just credentialism. Maybe a zillion other things.
> “My major was Physics and not Computer Science.”
Could be related to this. I know many good engineers who are better than average but are not CS majors, and who struggle with whiteboard/leetcode interviews.
The amusing thing is that people assume autonomous vehicles are a deep learning problem. So far, the pure deep learning people haven't done all that well at it. Most of what Waymo does is geometry and sensor fusion. Visual object recognition and future behavior of moving obstacles is used, too, but that's not the core of the system. Look at their videos, showing an above view of what's around the vehicle and "fences" showing where it shouldn't go.
29 comments
[ 7.0 ms ] story [ 75.4 ms ] threadBy-the-by, if you want to work on autonomous vehicles and aren't too attached to being neck deep in the deep learning parts, nearly all autonomous vehicles companies need an army of engineers for data handling, and on-vehicle performance.
(Vladimir, if you're reading this, I'd be curious to hear what your $$$ investment was for your kaggle competitions).
No wonder good people aspire to go to the US.
1) education is pretty good (Top 50-200)
2) engineers are cheap
What makes you say this?
I remember that moment to this day. I am standing at the scene. The organizer is preparing a check and some gifts. Alexey and I emerge victoriously, but I can’t help but feel frustrated by the apparent absurdity. How did this happen that some random dude that does not have domain knowledge in medical imaging got first places in both challenges? And yet, people that make money for a living working on medical imaging use much weaker models?
I asked the audience: “Do you know where do I work?” Noone, except one organizer who checked my LinkedIn profile knew. I told them that I work at TrueAccord, a debt collection agency and that I did not train deep learning models at work. I lamented that I was not able to break away from this model because HRs in Google Brain and Deepmind did not even look at my resume.
>During that same period, I had failed an onsite interview at Descartes Labs, and a few technical screens.
>Somewhere around that time, I was invited for the onsite interview with NVidia, which I also did not pass. One of the issues that I had was my limited knowledge of how 2D object detectors work.
>The next company on my list was Tesla. The recruiter contacted me because of my Kaggle achievements, which did not happen often. I passed take-home tests, tech screenings, and an onsite interview. The next steps were the background check and approval of my application by Elon Musk. I did not pass. [He'd violated a pretty bullshit sounding NDA]
>The company that was organizing the competition was called Planet Labs. They had an open DL Engineer position, I asked about it and was invited to an onsite interview. I failed again. The feedback — not in-depth DL knowledge.
He got a number of interviews, which he didn't pass. Is it really so surprising that Google Brain and Deepmind are going to be pickier than most?
Could be related to this. I know many good engineers who are better than average but are not CS majors, and who struggle with whiteboard/leetcode interviews.