This is an introduction to neural networks and related learning paradigms. This module provides an overview of backpropagation learning, and explores Pyro's neural network library, conx. When completed, the reader should be ready to design and create neural networks for learning on- and off-line with a robot.
Neural Networks in Pyro
To create neural networks in Pyro we will use the Conx module, which is a network scripting language and environment. Conx was designed to be used by connectionist researchers, as well as a teaching tool for AI and robotics courses. The idea behind this system is to allow experimenters to quickly and easily create, train, and test basic architectures such as feedforward and simple recurrent neural networks.
Conx is primarily based on three classes: Layer, Connection, and Network.
A Layer is a collection of nodes.
A Connection is a collection of weights linking two layers.
A Network is a collection of layers and connections.
A Network contains all of the parameters and methods necessary to run back-propagation learning.
Conx includes a specialized Network class, called SRN, which can be used to create an Elman-style simple recurrent network.