Should Hebbian learning be selective for negative excess kurtosis?

GROS, CLAUDIUS; ECKMANN, SAMUEL; ECHEVESTE, RODRIGO

Praga

Encuentro; 24th Annual Computational Neuroscience Meeting: CNS*2015; 2015

Within the Hebbian learning paradigm, synaptic plasticityresults in potentiation whenever pre- and postsynapticactivities are correlated, and in depression otherwise. Thisrequirement is however not sufficient to determine theprecise functional form for Hebbian learning, and a rangeof distinct formulations have been proposed hitherto.They differ, in particular, in the way runaway synapticgrowth is avoided; by either imposing a hard upper boundfor the synaptic strength, overall synaptic scaling, or addi-tive synaptic decay [1]. Here we propose [2] a multiplica-tive Hebbian learning rule which is, at the same time, self-limiting and selective for negative excess kurtosis (for thecase of symmetric input distributions).