Spike-Time Dependent Plasticity Learning Techniques for Event-Based Signals

Spiking Neural Networks (SNNs) are a family of more biologically plausible artificial neural networks, whose main advantage over classical ones is their energy consumption when tailored to the appropriate hardware. The big gap between the amazing results of AI and SNNs today is due to the inherent difficulties in training these particular models. In this thesis, several approaches to the training of SNNs are discussed and the main focus is on spike-time dependent plasticity, a technique belonging to the family of Hebbian learning rules. Several experiments are also conducted and aimed at gaining insights into the effectiveness of these training approaches.

Spiking Neural Networks (SNNs) are a family of more biologically plausible artificial neural networks, whose main advantage over classical ones is their energy consumption when tailored to the appropriate hardware. The big gap between the amazing results of AI and SNNs today is due to the inherent difficulties in training these particular models. In this thesis, several approaches to the training of SNNs are discussed and the main focus is on spike timing-dependent plasticity, a technique belonging to the family of Hebbian learning rules. Several experiments are also conducted and aimed at gaining insights into the effectiveness of these training approaches.