Machine Learning: The High-Interest Credit Card of Technical Debt
sazius, Charles Stanhope likes this.
clacke@libranet.de ❌, clacke@libranet.de ❌, clacke@libranet.de ❌, clacke@libranet.de ❌ and 1 others shared this.
Hmm, I should read that.
It may affect your field a bit less since you do computer vision more specifically? I dunno.
At any rate, the tl;dr is that machine learning allows getting a lot of functionality added very fast in many domains, but you don't necessarily know how it works, and if you need to "change things" you don't really know what consequences that's going to have, or might even have the right tools to hop in and make tweaks.
Christopher Allan Webber at 2017-05-03T17:21:45Z
clacke@libranet.de ❌ likes this.
Yeah, I certainly think there's a risk with using machine learning without understanding what it is doing, and its limitations. With all the hype these days a lot of people jump into it without necessary knowing all that. I did machine learning back before it was all the rage ;-)
Also it might be useful to distinguish deep learning and neural networks from things like probabilistic machine learning and kernel machines, which are easier to understand and interpret.
But it's also the "black box" nature of deep learning that is intriguing, and probably much of its power comes from that, in contrast to being hand-crafted rules that are easy to understand, but also fundamentally limited to just encoding expert knowledge and not learning anything new. Probably biological brains work in a similar "emergent" fashion, and one shouldn't dismiss the single working example of intelligence that we have :-)
sazius at 2017-05-03T17:33:52Z
Yutaka Niibe, clacke@libranet.de ❌, Charles Stanhope, Christopher Allan Webber likes this.
