Connectionism

• One of my pet peeves is that connectionists call their work neural networks.
• Connectionist systems use a number of relatively simple units inspired by the brain.
• Neural networks, in my terminology, are actual models of biological neural systems.
• A lot of connectionist work is relevant to machine learning.
• Multi-Layer perceptrons learning via back-propagation are a standard and general machine learning algorithm.
• RBFs, neural gas and other connectionst systems are also useful.
• Self-organising maps say something about the brain but not much.
• Hopfield nets are often used as neural models, but are Topologically unrealistic.
• Adaptive resonance theory is purported to be a neural model, but it is not at all clear how it correlates with neurons, or even if it does.