rockpool.devices.dynapse.quantization.autoencoder.loss

Dynap-SE autoencoder based quantization loss computation package

  • Non User Facing *

Functions

loss_reconstruction(encoder,Β parameters,Β input)

loss_reconstruction calculates the mean square error loss between output and the target, given a new parameter set.

penalty_negative(param)

penalty_negative applies a below zero limit violation penalty to any parameter

penalty_reconstruction(n_bits,Β bit_mask)

penalty_reconstruction applies a penalty if the bit_mask encoding&decoding is non-unique.