A Surrogate-Assisted Differential Evolution Approach for the Optimization of Ben’s Spiker Algorithm Parameters

Spiking neural networks (SNNs) differentiate themselves from traditional artificial neural networks by modeling the behavior of neurons in a more biologically plausible manner. Consequently, they employ discrete spikes or events to communicate information. Therefore, codifying analog signals into spike trains is a fundamental pre-processing step in SNNs. Ben’s Spiker Algorithm (BSA) has become one of the most used codification methods. Moreover, having optimal parameters allows for efficient and desirable performance. This paper contrasts two Kriging-Assisted Differential Evolution (KADE) approaches against Differential Evolution (DE) in said optimization task. The implementation is tested in a synthetic signal, and an electroencephalographic (EEG) signal to assess the consistency of the method. Furthermore, the Signal to Noise Ratio (SNR) metric was used to evaluate the performance of the implementations. Our findings demonstrate that KADE reduces the computational time of the implementation while achieving similar reconstructed signals. Specifically, the KADE approaches and DE implementation achieved a mean SNR of 9.04, 9.08, and 9.30, respectively, for the synthetic signal while reaching 10.27, 10.79, and 11.43, respectively, for the EEG signal.