Ask HN: How was your machine learning interview at Google/Facebook/Amazon
Looking for feedback from folks who applied for Machine Learning engineer or related positions at Google/Facebook/Amazon. How was your interview? How focused was the interview on ML topics as opposed to coding? How big was the emphasis on Deep Learning as opposed to more traditional methods? Was the interviews math heavy?
6 comments
I recently interviewed for a machine learning position at Google. There was essentially no focus on machine learning. I had 6 interview sessions total (5 onsite and 1 live coding phone screen) and 5 of those interviews had nothing to do with machine learning. In fact, those 5 interviewers had no machine learning background and knew nothing about machine learning. The one interview that was focused on machine learning was fairly straightforward and simple. It was kind of funny how large a contrast there was between the Google recruiter'a pitch ("we want to become an AI first company") and my actual experience interviewing. The interview process I went through has no way of distinguishing between someone with in depth machine learning knowledge and someone with basic machine learning knowledge. It was essentially a software engineer interview. I did not get the position, but I would recommend you look elsewhere if your focus is truly machine learning
>I did not get the position, but I would recommend you look elsewhere if your focus is truly machine learning
For ML, it's really important to join a team where your interests in ML are aligned with what the team does. It's really hard to see this from a job description. There's enough going on at Google that you can find work that fits.
I've interviewed twice at Google and had the same experience as you. No ML or math questions at all. More algorithms and how to quantify a business problem. That being said, I asked enough questions to realize that some groups use ML, but that's a small part of what they are doing. For example, they might have a platform for doing A/B testing and the "Data Scientist" job is really defining A and B and feeding that into the platform to extract metrics. How much ML being done is going to be different on the Ads team than a customer facing services role for Google Cloud.
I had similar experiences interviewing at Facebook, just with more probability and stats brain teasers. Facebook doesn't guarantee which job you'll get once you're in. You go to the bootcamp and then which team you end up is decided after the bootcamp. That doesn't work for everybody if there are certain types of ML work you're not interested in.
For ML, it's really important to join a team where your interests in ML are aligned with what the team does. It's really hard to see this from a job description. There's enough going on at Google that you can find work that fits.
I've interviewed twice at Google and had the same experience as you. No ML or math questions at all. More algorithms and how to quantify a business problem. That being said, I asked enough questions to realize that some groups use ML, but that's a small part of what they are doing. For example, they might have a platform for doing A/B testing and the "Data Scientist" job is really defining A and B and feeding that into the platform to extract metrics. How much ML being done is going to be different on the Ads team than a customer facing services role for Google Cloud.
I had similar experiences interviewing at Facebook, just with more probability and stats brain teasers. Facebook doesn't guarantee which job you'll get once you're in. You go to the bootcamp and then which team you end up is decided after the bootcamp. That doesn't work for everybody if there are certain types of ML work you're not interested in.
Wow that sounds like such a boring job.
For the the type of role I described in my post, being able effectively define metrics to quantify a business problem (e.g. how can we better engage with group X on our platform) and work with product engineering to build those features (including helping to define the data collected and how it's stored) into the platform is more important than tweaking a bunch of parameters in a machine learning model.
A lot of those folks are not only thinking of ways to quantify a business problem, they are actually thinking of new business problems (e.g. what does it even mean to "engage" on our platform). It can be quite creative and challenging.
Unless you are writing the core algorithms or working as a statistician, a lot of ML jobs are some variation of the above - basically coming up with ways to turn business problems into data/features you can feed into a model and picking an appropriate model. How much code you write and the tools/languages you are using will depend on the job and size of the company.
A lot of those folks are not only thinking of ways to quantify a business problem, they are actually thinking of new business problems (e.g. what does it even mean to "engage" on our platform). It can be quite creative and challenging.
Unless you are writing the core algorithms or working as a statistician, a lot of ML jobs are some variation of the above - basically coming up with ways to turn business problems into data/features you can feed into a model and picking an appropriate model. How much code you write and the tools/languages you are using will depend on the job and size of the company.
To be fair to Google, the position was titled "software engineer, machine learning", but the recruiter recommended that position for someone with my skill set
What were some of the questions they asked?