Item

Abstract

This paper suggests a learning-theoretic perspective on how synaptic plasticitybenefits global brain functioning. We introduce a model, the selectron, that (i)arises as the fast time constant limit of leaky integrate-and-fire neurons equippedwithspikingtimingdependentplasticity(STDP)and(ii)isamenabletotheoreticalanalysis. We show that the selectron encodes reward estimates into spikes and thatan error bound on spikes is controlled by a spiking margin and the sum of synapticweights. Moreover, the efficacy of spikes (their usefulness to other reward maxi-mizing selectrons) also depends on total synaptic strength. Finally, based on ouranalysis, we propose a regularized version of STDP, and show the regularizationimproves the robustness of neuronal learning when faced with multiple stimuli.