The Attraction Indian Buffet Distribution

We propose the attraction Indian buffet distribution (AIBD), a for binary feature matrices influenced by pairwise similarity information. Binary are used in Bayesian models to uncover latent variables (i.e., features) that explain observed data. The process (IBP) is popular exchangeable prior matrices. In presence of additional information, however, exchangeability assumption not reasonable or desirable. AIBD can incorporate yet it preserves many properties IBP, including total number features. Thus, much interpretation and intuition one has IBP directly carries over AIBD. A temperature parameter controls degree which information affects feature-sharing between observations. Unlike other nonexchangeable distributions allocations, probability mass function tractable normalizing constant, making posterior inference on hyperparameters straight-forward using standard MCMC methods. novel sampling algorithm proposed demonstrate feasibility as allocation compare performance competing methods simulations an application.