Hacker News Reader

Comment by chinston | Hacker News Reader

To ML/DL engineers out there: do I have to know the math to get a job as a ML/DL engineer?

I ask because all these tutorials (including Fast.ai) focus on eliminating the need to know the math. So I'm wondering, is the day-to-day work consist of utilizing the necessary libraries and letting them worry about the math?

View on HN

Well, you could get in Google without knowing any math, as long as you pass the interview full of traditional algorithm questions. Then you may get matched with a team who does ML work, if you have the resume for it. (This doesn't work if you want to be a research scientist at Google, though).

But I won't recommend such approach. ML almost never works on the first try. But it is unique among computer science in that many errors cannot be found by tests. You will just get a lower accuracy score, with no idea why. Without a deep understanding of the theory, you will have a hard time debugging.

not knowing the math behind ML is like not knowing a language and using google translate. most of the time you'll be fine but you won't be able to tell when something goes wrong and you won't be able to figure out why, and then you'll get things like

http://www.itltranslations.com/wp-content/uploads/2017/11/5_...

I don't think it's the right comparison. We use plenty of ml models in our startup like lsh and kmeans and I am hardly aware of how the math behind it runs.

If anything, these ml programs are supposed to abstract the maths behind it.

As a data scientist that heavily uses DL in my day-to-day, yes you need to know math. We are having difficulty hiring, and few candidates can answer basic questions: what is an FFT, what is the central limit theorem, what is an eigen value, what does the trace of a matrix tell you, etc. Many candidates use ML but have no idea what they are applying to the data.

Reply on news.ycombinator.com