implementations of the paper Learning under weight constraints in networks of temporal encoding spiking neurons

[w5688414]

Does anyone implement square cosine encoder for the input neurons of this paper:

Learning under weight constraints in networks of temporal encoding
spiking neurons

[SimonInParis]

I never read that. Sounds to me like an engineering solution to map the x values into cos^2(normalized(x) + random phase).
Why not use a simple normalized(x)^y, testing y in the range [0.1...5.0] to see if it fits better? I don't have any argument for the cosine function to perform better than any other mapping function. I understand the benefit of using a mapping function, though.

BTW I tried to play with Iris with very limited success.
But I agree that normalizing this kind of dataset is a good starting point.
Mapping it before converting to Poisson can prove beneficial (Poisson randomization being somehow equivalent to the random cosine phase across the input neurons).