I've read a bit about deep learning, and it seems quite compelling. Basically, we finally figured out how to train the kinds of feedback networks everyone was excited about in the 70s. And surprise, they're awesome. What are your thoughts? * Great for feature extraction/dimensionality reduction? * Great for supervised learning? * Great for unsupervised learning? * Interesting, but too finicky to rely on? * Effective, but difficult to intuit/trust so unsuitable for business applications?
I've felt all these things at various times about neural networks. They're neat, and when they work it's impressive, but I've had trouble in the past with them being extremely finicky about learning rates/configuration, and they seem potentially difficult to debug and I could have trouble trusting them when reliability really mattered.
[link][comment]
