Skip to main content
. 2022 Nov 11;16:951164. doi: 10.3389/fnins.2022.951164

Figure 1.

Figure 1

The workflow is summarized in five steps. Dataset acquisition and signal encoding (A) with analysis of information content and reconstruction loss (B). Different non-spiking classifiers are identified and employed to produce references for the proposed Recurrent Spiking Neural Network (RSNN) (C), with the latter undergoing a hyperparameter optimization (D). Finally, performances are evaluated, accounting for different metrics and hardware implementations (E).