Results

**1 - 1**of**1**### Non-Parametric Discrete Mixture Model Recovery via Nonnegative Matrix Factorization

"... Mixture modeling expresses probability densities as convex combinations of constituent probability distributions: qi(x) = r∑ j=1 wijpj(x), (1) Here qi and pj are density functions, and wij are nonnegative weights, summing to unity for each i. In classical mixture modeling, the constituent density fu ..."

Abstract
- Add to MetaCart

(Show Context)
Mixture modeling expresses probability densities as convex combinations of constituent probability distributions: qi(x) = r∑ j=1 wijpj(x), (1) Here qi and pj are density functions, and wij are nonnegative weights, summing to unity for each i. In classical mixture modeling, the constituent density functions, pj, are assumed to be from some class of parametric distributions. Various well established algorithms, typically using expectation minimiza-tion, can optimally recover the weights, wij, given an observed sample of values from the qi distributions [1]. In certain settings, however, mixture modeling is desirable, but the constituent distributions are neither known in advance,