People have tried using probabilistic SNNs because the gradient is well defined there. The issue is it’s computationally intractable to calculate so you still h

People have tried using probabilistic SNNs because the gradient is well defined there. The issue is it’s computationally intractable to calculate so you still have to use an approximation and you’re thus not better off than the people using a surrogate gradient of the non-probabilistic SNN.