Fig. 8: GLE for challenging spatio-temporal classification problems. | Nature Communications

Fig. 8: GLE for challenging spatio-temporal classification problems.

From: Backpropagation through space, time and the brain

Fig. 8

Averages and standard deviations measured over ten seeds. Top row: samples from the a MNIST-1D, b GSC—including raw and preprocessed input—and c CIFAR-10 datasets. d Performance of various architectures on MNIST-1D. Here, we used a higher temporal resolution for the input than in the original reference60. e Performance of various architectures on GSC. f Performance of a (G)LE LagNet architecture on MNIST-1D. For reference, we also show the original results from ref. 60 denoted with the index 0. g Performance of a (G)LE convolutional network on CIFAR-10 (taken from ref. 17) and comparison with BP.

Back to article page