Stumpokapow
listen to the mad man
I'm actually not sure you need much of the math -- you probably want concepts that are mathematical in nature (having a basic understanding of bias/variance tradeoff and thus overfitting; getting the basic gist of generative vs. discriminative classifiers; understanding at a STAT 102 type level what regression and classification are).
But, like, most of the techniques you use are canned implementations that are easy to implement without knowing the math. Need to do PCA in R? prcomp is one line. Even if you don't understand singular value decomposition. Need to run elastic net? cv.glmnet into glmnet. Does it matter if you can't solve the Lagrangian dual or even understand what convex optimization is? I don't think so.
I mean, basic literacy isn't going to hurt anyone, of course. If you run a method and it fails due to having a rank deficient data matrix, then understanding that the technique requires inverting the data matrix and doesn't work on p>n problems is useful. But then again, if you google the error message, you'll find a stackoverflow comment that explains that even if you don't know any linear algebra.
This response doesn't apply to hiring standards or job expectations -- I'm in a very very very very different part of the ML landscape from bioinformatics -- but rather that I expect a programmer with even baseline competence could become proficient enough in ML techniques to use them professionally without a math background.
