Extended Data Fig. 3: Image classification details.
From: Deep learning incorporating biologically inspired neural dynamics and in-memory computing
a, Complete spiking CNN architecture. b, CNN learning curve for rate-coded inputs without preprocessing. The accuracy was calculated by averaging over ten different initializations (vertical brackets) or over the last 50 epochs (horizontal brackets). c, Analogous CNN learning curve for rate-coded inputs obtained from MNIST images preprocessed with elastic distortions. d, Table comparing the state-of-the-art fully-connected (FC) and convolutional (CNN) SNN architectures27,46,49,53,57,87,88 in terms of parameters and obtained MNIST accuracy.
