Dendrites decrease the synaptic weight resolution necessary to compute

Abstract

In theory, neurons can compute all threshold functions, but in practice, synaptic weight resolution limits their computing capacity. Here, we study how dendrites alleviate this practical limitation by demonstrating a computation where dendrites considerably decrease the necessary synaptic weights resolution. We show how a biophysical neuron model with two passive dendrites and a soma that can implement this computation more efficiently than a point neuron. The latter requires synaptic weight orders of magnitudes larger than the others, the former implement the computation with equivalent synaptic weights. This work paves the way for a new generation of neuromorphic chips composed of dendritic neurons. These chips will require less space and less energy to function.