5 ms·

Shifting Careers to Autonomous Vehicles

104 points by dsr12 7y ago

niggers 7y ago
niggers
thathoo 7y ago
Kudos to your effort and resilience even in the face of failures. You faced your challenges head on and took the bull by the horns! Love it!
lostdog 7y ago
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).
bitL 7y ago
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).
- duaoebg 7y ago
  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.
  - systemtest 7y ago
    Here in the Netherlands they will pay competitively... for management positions. Even middle management gets compensated very well. Engineers not so much.
    - himlion 7y ago
      Agreed, contracting is the way to go if you want decent money.
    - readyoubestbook 7y ago
      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)?
  - arvinsim 7y ago
    I think it is the same with Singapore too. No wonder good people aspire to go to the US.
  - Enginerrrd 7y ago
    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.
- nicoburns 7y ago
  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).
2sk21 7y ago
Very impressed how the author was able to bootstrap your way into a good job.
thecleaner 7y ago
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.
realbarack 7y ago
This is an inspiring story. Nice work and well-earned!
ct0 7y ago
This is inspiring! Excellent work and write up. Thanks for sharing.
KKKKkkkk1 7y ago
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.
- 6gvONxR4sf7o 7y ago
  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?
  - OnlineGladiator 7y ago
    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.
Animats 7y ago
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.

104 points by dsr12 7y ago

niggers

Kudos to your effort and resilience even in the face of failures. You faced your challenges head on and took the bull by the horns! Love it!

lostdog 7y ago

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).

bitL 7y ago

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).

duaoebg 7y ago
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.
- systemtest 7y ago
  Here in the Netherlands they will pay competitively... for management positions. Even middle management gets compensated very well. Engineers not so much.
  - himlion 7y ago
    Agreed, contracting is the way to go if you want decent money.
  - readyoubestbook 7y ago
    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)?
- arvinsim 7y ago
  I think it is the same with Singapore too. No wonder good people aspire to go to the US.
- Enginerrrd 7y ago
  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.
nicoburns 7y ago
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).

2sk21 7y ago

Very impressed how the author was able to bootstrap your way into a good job.

thecleaner 7y ago

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.

realbarack 7y ago

This is an inspiring story. Nice work and well-earned!

ct0 7y ago

This is inspiring! Excellent work and write up. Thanks for sharing.

KKKKkkkk1 7y ago

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.

6gvONxR4sf7o 7y ago
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?
- OnlineGladiator 7y ago
  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.

Animats 7y ago

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.