US 11,983,620 B2
Simplification of spiking neural network models
Henry Markram, Lausanne (CH); Wulfram Gerstner, Ecublens VD (CH); Marc-Oliver Gewaltig, Ecublens VD (CH); Christian Rössert, Lucerne (CH); Eilif Benjamin Muller, Preverenges (CH); Christian Pozzorini, Lausanne (CH); Idan Segev, Jerusalem (IL); James Gonzalo King, Lausanne (CH); Csaba Erö, Ecublens VD (CH); and Willem Wybo, Prilly (CH)
Assigned to Ecole Polytechnique Federale De Lausanne (EPFL), Lausanne (CH)
Filed by Ecole Polytechnique Federale De Lausanne (EPFL), Lausanne (CH)
Filed on Apr. 8, 2022, as Appl. No. 17/658,462.
Application 17/658,462 is a continuation of application No. 15/942,765, filed on Apr. 2, 2018, granted, now 11,301,750.
Claims priority of provisional application 62/479,415, filed on Mar. 31, 2017.
Prior Publication US 2022/0230052 A1, Jul. 21, 2022
Int. Cl. G06N 3/04 (2023.01); G06N 3/049 (2023.01); G06N 3/08 (2023.01); G06N 3/082 (2023.01)
CPC G06N 3/049 (2013.01) [G06N 3/08 (2013.01); G06N 3/082 (2013.01)] 18 Claims
OG exemplary drawing
 
1. A method for simplifying a spiking neural network model, the method comprising:
providing the spiking neural network model to be simplified, the spiking neural network model comprising a first neuron having an arborized projection that connects to a second neuron and to a third neuron, wherein the arborized projection is to convey input from the second neuron and the third neuron to the first neuron;
defining a first temporal filter for the conveyance of input from the second neuron to the first neuron along the arborized projection;
defining a second temporal filter for the conveyance of input from the third neuron to the first neuron along the arborized projection, wherein the first temporal filter differs from the second temporal filter;
replacing, in the spiking neural network model, the arborized projection with
i) a first connection extending between the first neuron and the second neuron, wherein the first connection filters input from the second neuron in accordance with the first temporal filter and
ii) a second connection extending between the first neuron and the third neuron, wherein the second connection filters input from the third neuron in accordance with the second temporal filter;
fitting
a) behavior of the spiking neural network model in which the arborized projection is replaced
to
b) behavior of the spiking neural network model to be simplified;
after the fitting of the behavior, comparing
c) behavior of the spiking neural network model in which the arborized projection is replaced
to
d) behavior of the spiking neural network model to be simplified over a duration of time; and
amending the fitting of the behavior of the spiking neural network model in which the arborized projection is replaced based on the comparison.