Hacker News Reader

Comment by itg | Hacker News Reader

Good luck with that. Any place doing serious ML will require the person to have a PhD or have publications and presentations at conferences like NIPS/ICML. Even most CS grads with a bachelors do not have the math background required unless they double majored in math or stats.

This is more VC/founders who are hyping up AI and need more ML folks so they can drive down costs.

View on HN

The notion that you can only do "serious ML" if you have a PhD/published papers is a No True Scotsman fallacy.

The modern tools for ML/deep learning are accessible to all open source and well documented. And as I note in my top-level comment, old-fashioned statistical methods like linear regression are more than sufficient for real-world business problems, and definitely do not require a PhD to grok.

This is what I keep telling people too - that the "old-fashioned" statistical methods aren't applied to more than a tiny little fraction of the problems they could be applied to yet.

I didn't double major in math or stats. I squeaked through linear algebra without understanding the material, and then re-studied it independently a few years ago and actually understood a little more.

Yet, I can still apply machine learning to solve Ax=b problems. More importantly, I can use business analysis and write code to transform business problems into an Ax=b problem, and then optimize it.

You don't need a PhD to grok optimizing a vector to transform a matrix of inputs into a vector of observed outputs, then apply that trained vector going forward. Neural nets are slightly more complicated than a straight linear regression, but only slightly. I'd call decision tree methods like GBM even more complex, but still eminently grokkable for a decent programmer.

> Any place doing serious ML will require the person to have a PhD or have publications and presentations at conferences like NIPS/ICML.

Stop spouting this bullshit. You don't need a PhD, and you don't need to advance the field to be doing 'serious ML'. All you need to be able to do is know how and when to apply it to solve crucial business problems.

This misunderstanding is what needs to go the way of the dinosaur. You don't need a PhD to know how to clean data, prepare features, or generate synthetic features. You don't need a PhD to tune hyperparameters. You don't need a PhD to know how to separate your test data from your training data. You don't even need a PhD to wire things together at a high level with something like tensorflow.

Obviously you will want a theoretical expert on your team. But if your startup is counting on having a room full of them, good luck.

>Any place doing serious ML

Good thing the vast majority of businesses won't need "serious" ML, but instead will require only simple implementations to help solve business problems.

Utter nonsense. A PhD signals two things: that a person has the same degree of mastery of core material as a person with a master's degree and that he or she has the determination to do additional original research sufficient to produce a 100 page paper.

It isn't required for any serious research effort, except by the accident of inertia. And it certainly isn't a necessary indicator of determination.

You seem to make extremely light of "doing original research" here. A PhD or equivalent is absolutely required to do serious research in the field. Sure, you can always get the depth of knowledge required without a formal program, but the reality is that few do. Most people just take Machine Learning 101 then think they are a domain expert.

That said, there is definitely a place for non-PhD level ML practitioners. I don't think the industry has stabilized in this regard, but I can definitely see a "machine learning developer" type position becoming quite common. This is not the same as someone doing original research, but would definitely meet the needs of a great many business use cases.

I'm not taking it lightly at all, considering I've done it. I know exactly what it requires and what it signals.

> Utter nonsense.

This counts as name-calling in the sense that the HN guidelines ask you not to do it: https://news.ycombinator.com/newsguidelines.html. Your comment would be much better without that bit.

We detached this subthread from https://news.ycombinator.com/item?id=13599533 and marked it off-topic.

Reply on news.ycombinator.com