The Third Generation of Neural Networks: Beyond simple nodes and linear synapses

Networks: Beyond simple nodes and linear synapses Jaap Hoekstra and Arthur van Roermund Delft University of Technology, Faculty of Electrical Engineering/DIMES P.O. Box 5031, 2600 GA Delft, The Netherlands tel. +31 (0)15 278 3836 fax. +31 (0)15 278 5922 [email protected] Abstract|The paper discusses an analogue neural network concept in which the neuron is split up into parts (more precise, we consider the dendrite as a seperate and subdivided entity) and the synapse is non-linear. The neural network constructed in this way belongs to the third generation of neural networks. Keywords| (Arti cial) Neural Networks, Neural Computing, Dendritic Computation, Biologicinspired subsystems, Nano-scale electronics, Learning, Compartmental modeling.