<div class="bbWrapper">I've had a successful startup exit thanks to a recommendation system I had developed. I <b>do not</b> recommend this subfield of machine learning, as the value you are generating is limited. It is great if you work for Amazon, Netflix, and Spotify, but very low for just about everybody else, including Facebook, Google, Linkedin, and Microsoft. It is not that the value of recommendations is low, the value of <b>better</b> recommendations is low. (You may think that the recommendations provide a conversion lift, but very often they convert people who would've converted anyway. Imagine a recommendation system recommending posts on this forum. Useful? Yes. Critical? No. If you are building system, by the Law of Effection, your systems have to be where the money is. Your ML has to be mission-critical for the business to make money, if you want to have upside and to be treated well. ) <br />
Besides, the skills are poorly transferable. It is less rigorous than image processing, less lucrative than creating trading strategies; you don't learn as much about marketing as you would from directly optimizing funnels and A/B testing campaigns. It's just a subfield offering limited opportunities, in my view.<br />
<br />
Besides, Netflix admitted at a conference that the metrics used "do not allow to distinguish between an enhanced quality of life and an addiction." I've asked a bunch of people at a RecSys conference, and everyone building recommendation systems admitted that they are basically trying to get people addicted to the Internet, wasting their lives online. Intentionally. How's that for a sense of purpose?<br />
<br />
<br />
That aside, if you are good, you can join an early-stage ML startup and get an upside in the form of stock/ equity. Or you can start one. If you are not an expert, become one --- or forget it. If you want to go that way, PM me, send me your resume, and I might have some advice.<br />
<br />
And there are sooooo many biggest problems when it comes to the data, I can't even begin. There's collecting this data. There's accessing it. There's finding competent tech people who can do something with it. Then there's finding competent business people who can decide what to do with that, because tech people don't care about $. Then... don't get me started.<br />
<br />
In terms of predicting stock performance, it's a bit harder than that. Here is what it's like: you've had a trading model that used to work, then it suddenly doesn't work and starts losing money. You spend 10-12 hour days at work, trying to figure out how to fix it, before you lose your bonus if your strategy does not start performing again asap, and then your job, and the efforts of several last years with that. You need a solid tech background, AND THEN you need to commit another 5 years of overtime work if you want to get real results creating trading strategies that make money.<br />
I've rejected it for myself essentially because it's a SlowLane, and if you fail, you get crappy skills out of it. If you fail to get rich as a sales person, at least you get certain social skills. If you fail at creating trading strategies, your last 5 years of ultra-specialization are of little use, unless you want to spend the rest of your career doing something similar. And some specialize even more, by fine-tuning the strategies to custom hardware. Even owning a trading firm is NOT a Fastlane, it's a lifestyle.</div>