Sir,
I use the clip-shift-floor to convert a network to SNN in a regression problem. I discover a phenomenon that the MSE drop to the lowest point when T=L and then rise to a regular value which is higher than ANN's loss.
( e.g. ANN validate loss = 0.03 under L = 8, SNN validate loss = 0.041 when T = 128 but SNN lowest loss = 0.028 when T = 8 = L)
I don't understand it as you do, please enlighten me.
Expecting for your reply.
Sir, I use the clip-shift-floor to convert a network to SNN in a regression problem. I discover a phenomenon that the MSE drop to the lowest point when T=L and then rise to a regular value which is higher than ANN's loss. ( e.g. ANN validate loss = 0.03 under L = 8, SNN validate loss = 0.041 when T = 128 but SNN lowest loss = 0.028 when T = 8 = L) I don't understand it as you do, please enlighten me. Expecting for your reply.