The HTM sequence memory model. (A) The cortex is organized into six cellular layers.
Each cellular layer consists of a set of minicolumns, with each minicolumn containing
multiple cells. (B) An HTM neuron (left) has three distinct dendritic integration
zones, corresponding to different parts of the dendritic tree of pyramidal neurons
(right). An HTM neuron models dendrites and NMDA spikes as an array of coincident
detectors each with a set of synapses. The coactivation of a set of synapses on a
distal dendrite will cause an NMDA spike and depolarize the soma (predicted state).
(C, D) Learning high-order Markov sequences with shared sub-sequences (ABCD versus
XBCY). Each sequence element invokes a sparse set of minicolumns due to intercolumn
inhibition. (C) Prior to learning the sequences all the cells in a minicolumn become
active. (D) After learning, cells that are depolarized through lateral connections
become active faster and prevent other cells in the same column from firing through
intracolumn inhibition. The model maintains two simultaneous representations: one at
the minicolumn level representing the current feedforward input and the other at the
individual cell level representing the context of the input. Because different cells
respond to C in the two sequences (C’ and C”), they can invoke the correct high-order
prediction of either D or Y.
This site uses cookies. By continuing to use our website, you are agreeing to our privacy policy. No content on this site may be used to train artificial intelligence systems without permission in writing from the MIT Press.