1: \begin{abstract}

2:   We present the \textit{discrete infinite logistic normal}

3:   distribution (DILN), a Bayesian nonparametric prior for mixed

4:   membership models. DILN generalizes the hierarchical Dirichlet

5:   process (HDP) to model correlation structure between the weights of

6:   the atoms at the group level.  We derive a representation of DILN as

7:   a normalized collection of gamma-distributed random variables and

8:   study its statistical properties.  We derive a variational inference

9:   algorithm for approximate posterior inference.  We apply DILN to

10:   topic modeling of documents and study its empirical performance on

11:   four corpora, comparing performance with the HDP and the correlated

12:   topic model (CTM). To compute with large-scale data, we also develop

13:   a stochastic variational inference algorithm for DILN and compare

14:   with similar algorithms for HDP and LDA on a collection of $350,000$

15:   articles from \emph{Nature}.

16: \end{abstract}

17: