Clusteron

The learning rule used for both cell types was similar to the "clusteron" learning rule described in (Mel, 1992a), and involved two mechanisms known to contribute to neural development:

The total activation value of a clusteron is computed according to the observation that it behaves as an approximative low-order polynomial function.

These product terms were then summed across the dendritic tree as a whole, so that the clusteron was in effect a constrained sigma-pi unit with product terms for every pair of nearby synapses.