Implementation of SpikingGCN-N

[upperr]

Hello! Thanks for the source code of SpikingGCN. I have a question about the variant SpikingGCN-N, which uses the refined Heaviside activation function given in Eq. (8). I wonder how is this three-stage step function approximated to be differential for back-propagation, as the sigmoid function given in Eq. (10) seems only able to approximate the normal activation function in Eq. (7).

By the way, where can I find the code of SpikingGCN-N (if it is open-sourced)?

[Zihao0]

Hello, do you know how the Bernoulli encoder in the article is implemented? (Eq.(5) and Eq(10))

[Zihao0]

Sorry, I already know. It's Poisson encoder. Ha-ha