A Stigmergic Perceptron (SP) is a soft classifier of a time series with respect to a collection of archetypal time series, each representing different behavioral classes. The SP takes as an input a time series and provides as an output the linear combination of similarity values of the time series w.r.t. the most similar archetypal time series. Indeed the SP is organized, like a neural perceptron, in terms of linear combination of values provided by Stigmergic Receptive Fields (SRF). SRF can be combined and arranged into multilayer architectures thanks to specific input/output interfaces, namely clumping and activation. Each SRF provides a similarity measure between the input and the archetype time series. The term stigmergic refers to the internal logic of a SRF, inspired by stigmergy, an insects' coordination mechanism. Stigmergy is based on the spatio-temporal aggregation of input samples by functional structures called trails. Trails are generated by marks released by each sample in the input dimension, and evaporated over time. At the core of a SRF, the similarity between the trails of two time series is calculated. SP and the related SRFs are parametrically adapted to a given training set by means of the Differential Evolution (DE) algorithm.
Mario G.C.A. Cimino (2020). Stigmergic Perceptron (https://www.mathworks.com/matlabcentral/fileexchange/66526-stigmergic-perceptron), MATLAB Central File Exchange. Retrieved .
Architectural picture added