Terminology
Module: various encapsulated data processing algorithms which at minimum have an input and an output buffer in a forward implementation depending on free parameters usually adjusted by a machine learning algorithm. Also included are error buffers used in error back-propagation algorithms.
Network: objects consisting of modules which communicate via connections which may have adjustable parameters themselves. The network is itself a module.
Connections: objects that connect the layers of modules in the network.
Trainer: an object that uses a dataset to adjust the free parameters of the network.
Dataset: a set, typically very large, of inputs and desired outputs.
Network: objects consisting of modules which communicate via connections which may have adjustable parameters themselves. The network is itself a module.
Connections: objects that connect the layers of modules in the network.
Trainer: an object that uses a dataset to adjust the free parameters of the network.
Dataset: a set, typically very large, of inputs and desired outputs.
More to the concepts
Networks can be connected any number of ways as long as it results in a directed acyclic graph. As I understand this, once information leaves a layer it can't come back to that layer. Another crude way of saying it is no loops. Loops are not the same as error back-propagation.
How this encourages me
So let's talk about what I stated as an idea for learning about Neural Networks. I would like to create a small system of distributed nodes on a set of raspberry pi computers. According to the PyBrain documentation, as long as the arrangement of the networks results in a directed acyclic graph, I'm good to go. I could have a node that processes sound, one that processes camera input, and one that puts that information together to make decisions about a situation. I'm not saying that I know enough to do this yet; however, I am encouraged by the documentation to continue looking into the subject.
Brief question: Does anyone remember linear algebra? I need a refresher. Guess it's time for Khan Academy.
Brief question: Does anyone remember linear algebra? I need a refresher. Guess it's time for Khan Academy.
