Bayesian Inference for Nonnegative Matrix Factorisation Models

Figure 2

(Left) The family of densities $p(v;a\mathrm{,}b=1)=𝒢(v;a\mathrm{,}b/a)$ with the same mean $\langle v\rangle =b=1$. Small values of $a$ (for $a<1$) enforce sparser representations, and large values of $a\approx 100$ tie all values to be close to a nonzero mean (nonsparse representation).