Summary: LEARNING PRECISE SPIKE TIMES IN A TWO-VARIABLE SPIKING
Sven E. Anderson
Computer Science Program
Annandale, NY, U.S.A.
We evaluate the ability of reinforcement comparison learn-
ing to induce multispike patterns with sub-millisecond
precision in a two-variable spiking neural model. We
assume that a single reinforcement signal derived from
the fit of the produced spike pattern with a target pattern
is communicated with the neural model following pro-
duction of all spikes of the pattern. We find that arbitrary
multispike patterns can be learned with a precision of 0.2
msec. Patterns of one to five spikes can be learned with a
probability of success ranging from 20% to 70%.
pulse-coupled neural network, spiking, reinforcement