Connectionist Systems

• I'm not sure if I'm biased, but I think the most popular machine learning algorithms are connectionist.
• Multi-Layer Perceptrons with backpropagation is the most common. You can learn any function.
• Here you have layers of perceptrons. These operate on the vector of inputs and pass through the value. They are connected by weights. These get modified in a supervised manner using backpropagation of error.
• Self Organising Maps are also popular and are good for dimension reduction and unsupervised clustering.
• There are a range of other connectionist systems including Radial Basis Functions, Adaptive Resonance Theory, Neural Gas and a range of recurrent nets.