Demixing smells — fast inference in olfaction Supplementary Material

Because of the delta-function in the prior, performing efficient variational inference in our model is, as far as we know, difficult. Therefore, we smooth the delta-function, and replace it with a Gamma distribution, δ(cj) −→ Γ(cj |α0, β0). In the limit α0 → 0 and β0 → ∞ we recover the true prior. However, when these two parameters are finite, absent odors – odors with sj = 0 – have non-zero concentrations. To correct for this, we choose α0 and β0 so that there is an effective background