Spiking Neural Networks as a Machine Learning model have recently received a lot of attention as a potentially more energy-efficient alternative to conventional Artificial Neural Networks. Due to the non-differentiability and sparsity of the spiking mechanism, these models are not only expensive but also very difficult to train with algorithms based on propagating gradients through the spiking non-linearity. In this Thesis, we aim to address these two problems by developing a gradient-free training algorithm for Spike-Response-Model Spiking Neural Networks based on sampling and the solution of linear problems. To this end, we propose a fast, high-performance and interpretable algorithm inspired by the SWIM algorithm for Artificial Neural Networks, which we coin SNN-SWIM. This Thesis is comprised of a thorough theoretical discussion of the proposed scheme and supplementary numerical experiments.     «

Spiking Neural Networks as a Machine Learning model have recently received a lot of attention as a potentially more energy-efficient alternative to conventional Artificial Neural Networks. Due to the non-differentiability and sparsity of the spiking mechanism, these models are not only expensive but also very difficult to train with algorithms based on propagating gradients through the spiking non-linearity. In this Thesis, we aim to address these two problems by developing a gradient-free tra...     »