Layering

If a family of neurons sharing the same input field from some part of the sensorium may be crudely modelled as a dynamical and sequential form of a gaussian mixture model finding clusters in the data, we have some interesting grounds for speculation about statistical models. I shall avoid this and go straight to the propensity for neurons to form layers, and for the output of one layer to be the input to the succeeding layer. If each layer is accomplishing a model for the local clustering of its input, is a sequence of layers of such families implementing Syntactic Pattern Recognition? It is easy to see that this is not too hard a question to answer in the affirmative. The process of chunking I have described is readily seen to be easily implementable in layers of cluster finding entitities. The details need rather more formalism than I allow myself in this present book, but I claim it can be done. Moreover, the evolutionary process of layer formation can easily be accounted for along the lines discussed earlier, when I pointed out the variation in data that could be accommodated more effectively by having intermediate levels of UpWrite. We can essentially identify the levels of UpWrite with the layers of processing elements.