From what I remember from Chris’ course, neurons are noisy, so decoding with noise is a good idea. However, in the Nengo neural simulator, I’m confused about where noise comes into play. Here’s what I think I understand so far:
- There is no noise coming from the built-in neuron types, their behaviour is deterministic.
- Noise comes from built-up errors in approximation in a model.
- Noise can be taken into account explicitly by using the
LstsqNoise
andLstqMultNoise
solver or implicitly by regularisation.
Is this correct? Does noise come into play in other places in the reference Nengo simulator? If yes, how is it dealt with?