Subject: Re: [nupic-dev] Inter-layer plumbing, NLP
Hi, Jeff, et. al., I want to speak about the high order prediction and time classifier. I think this is equivalent to the propagation of excitation in the dendritic tree. Physiological STDP synapse plasticity model has a high temporal sensitivity, which can be used to explain the long time processes in the dendritic tree. The idea of a temporal matching in the dendritic tree come from the 60's, on my view (Radchenko,1968). My model of a dynamic associative memory is based on neuronal dendritic trees with depth of 20 segments that works as a shift register. The network from 500 neurons stores and reproduces from part several sentences. I have checked up to 60 symbols. Now the network from 2000 neurons in the 4 layers provides first-order logic with variables on the NLP example. Here we move from recognition to the processor. I see this model as part of the cortex column - the third pyramidal layer and around neurons. It is interesting to consider the place of these columns in HTM within the relation data - metadata. Selection of a word in a sentence has a lot of aspects from the pragmatics and up to psychology. In this regard, there is a question about the possible modeling purposes: - Model for the study of physiology - Markram . - Model for the study of mathematical ideas – McCulloch-Pitts neuron, perceptron. - Model for the construction of AI processor like the brain - HTM. We work in the last paradigm, which allows us to test progressive ideas, such as metabotropic synapses … DNA machine. There is a huge collection of ideas for the development and integration of models and solve the main problem. Ivan
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
