-6

I am very new to neural networks and only a lowly programmer. I don't have a firm grasp of the different neural network architectures. My question is this: what is the smartest architecture? Which network is the fastest learning, can recognize the most complex and vague patterns and is the most adaptable. I've been reading about all sorts of cool networks like the echo state and liquid state machines and long short-term memory networks but I really have no clue about how these work or which to use in what context. If you know how these crazy networks work I'd like to hear your thoughts.

thanks!

Aidan Collins
  • 1
  • 2
  • 3
    A proper answer to your question would have the size of a book, I'm afraid. It's too hard to compare all these networks without the reader having a certain ground knowledge. Diving into Neural Networks takes a long time to which there is no shortcut - but it's worth it. – runDOSrun Jul 04 '15 at 12:17

1 Answers1

2

There is no such thing that you seek. There are lots of different types of networks because there are lots of different types of problems and no one knows that the "best" way to do something is. Its an active area of research that is moving very quickly. There are certain types of networks that are becoming common for certain tasks (i.e.: convolutional networks used for image classification), but even those are not fixed in stone as the best. And even if they were, that tells you nothing about how many layers or how wide the layers need to be, and this doesn't even get into the initialization or updating of networks in training.

If you want to actually know and understand more about neural networks, you are going to have to do a lot of self study - it takes work and patience. A simple Q&A format like stack overflow will not help you understand much of anything about neural networks.

Raff.Edward
  • 6,404
  • 24
  • 34
  • Apparently online courses (Coursera/Youtube) are a good option for a lot of people. Considering that there aren't *that* much extensive modern ANN/Deep Learning books out there *yet*, it might be a good starting point as well (with same amount of work and patience ofc). – runDOSrun Jul 04 '15 at 12:19