Connectionism is a theory of information processing developed by Rumelhart & McLelland which models learning by the parallel processing of sub-symbolic systems.  Connectionist models represent the neurophysiology of the brain and use statistical properties.

Connectionist theory is therefore very different from early cognitive models in artificial intelligence, which are based on the serial manipulation of symbols via logical rules (e.g., Newell & Simon).

The basic unit of connectionism is the neuron:

It is an input device, receiving signals from the environment or other neurons.  It is an integrative device integrating and manipulating the input.  It is a conductive device conducting the integrated information over distances.  It is an output device sending information to other neurons or cells.  It is a computational device mapping one type of information into another.  And, it is a representational device subserving the formation of internal representations.

(Medler, 1998)

In all connectionist models, these units have connectivity, can be activated, have learning procedures which modify connections between units, and allow semantic interpretation of the network.

Activities

1. Read and reflect: Neural networks
2. Play: Activation