Check noise model in additive paper and possibly matlab code to use the same here

[duvenaud]

I checked, and I ended up using Gaussian noise and exact inference in the additive paper, so our results will be comparable without modification.