Sign up FAST! Login

Deep Learning 101

Stashed in: Software!, For Milo, Machine Learning

To save this post, select a stash from drop-down menu or type in a new one:

A nice overview of deep learning and what it has to offer. I expect we'll see a lot more of this as it matures. The achilles heel of supervised machine learning is usually "feature engineering" and one of the more compelling reasons that deep learning has so much potential is that it significantly eases so much of that feature engineering by more "automagically" discovering the most meaningful features.

Any feel for how mature the field of deep learning is? I mean, are these well-tested techniques?

It's one of the newer/sexier things in machine learning, but it's foundations are deep/solid and from a certain interpretation are rooted in neural networks, though it's a little bit more complicated than that. In my opinion, this is a type of learning that has some fairly broad and universal potential because of how it mitigates some of the feature engineering pain -- the kind of potential that may lend itself to being something like "machine learning in a can" where that can could be some fairly simple library calls that a typical developer could use without having to really understand what's going on. 

Although that can always be a little bit dangerous, that's the level that ML needs to get before it's going to *really* change the world. In the same way that most people have no idea about all of the stuff that happens in between making a request to a web server and getting back a response (nor should they really have to 99.9% of the time), I think that ML needs to get to that point of usability and general applicability that the mainstream can use it in a similar way for certain kinds of problems. When creative people can apply ML without having to be computer scientists or mathematicians, then we're going to see some truly amazing things start to happen around us.