MultiGaussian

Contents

MultiGaussian#

class sinabs.activation.MultiGaussian(mu: float = 0.0, sigma: float = 0.5, h: float = 0.15, s: float = 6, grad_scale: float = 1.0)[source]#

Surrogate gradient as defined in Yin et al., 2021.

https://www.biorxiv.org/content/10.1101/2021.03.22.436372v2

Parameters:

  • mu (float) – The mean of the Gaussian.

  • sigma (float) – The standard deviation of the Gaussian.

  • h (float) – Controls the magnitude of the negative parts of the kernel.

  • s (float) – Controls the width of the negative parts of the kernel.

  • grad_scale (float) – Scale the gradients arbitrarily.

Contents