A neural network consists of nodes and connections, where inputs flow to outputs to answer questions. A classic question is "what digit is this hand-written representation most likely to be?". Training a neural network requires flow in the opposite direction, updating the components in the network to answer the question.
Following Andrej Kaparthy's "Hacker's Guide to Neural Networks" in (def shef 34), we got as far as constructing the network, which led to interesting questions about how to deal with the training problem in a functional way. Solutions like the JavaScript implementation in Andrej's post rely on mutable objects in the network for the training operation. Start at section heading "Example: Single Neuron" to understand that solution.
This session will focus on the question of how to deal with updating networks like this in a functional style. Anyone wanting to look at any other aspects of neural networks is also welcome to come along and participate!