Artificial Neural Networks

Howdy folks,

I've been following the progress of ANN's lately, and it seems like every month there's a big innovation. The ability to generically create solutions for complex problems is great, but the abilities it enables for companies who have tons of real-world data is huge.

Since neural networks seem to be running behind almost everything nowadays, I think it's important for people to understand the potential capabilities. There's a newsletter and archive called Import AI which releases news about ANN's frequently : https://jack-clark.net/import-ai/

There's also https://arxiv.org/ where you can find new papers on neural networks and more.

Paul Werbos (the man who discovered backpropagation) has some interesting insights on how things are progressing in the area, and it seems he has concerns about companies wanting to push brain-computer interfaces to users. They're already starting to map human brain activity to ANN's to produce better image classification.

I feel these technologies are turning out to be the most pervasive to freedom and privacy, where it doesn't even require people anymore to, say, identify multiple objects in a picture or video, or detect emotional context in audio. For example, Facebook just recently implemented automated suicide prediction using ANN's, and that's just a feature for users.

There's plenty of good that can be done with ANN's too, and thankfully we have free software to make some pretty cool stuff that doesn't bite us at the same time. If anyone wants a simple feed-forward neural network to play with, I've got an implementation in c++ which I'd gladly share (note, not hardware-accelerated).

End of rant :D