Bayesian Sparse Unsupervised Learning for Probit Models of Binary Data
Ari Pakman, Ben Shababo, Liam Paninski · 2014
We present a unified approach to unsuper-vised Bayesian learning of factor models for binary data with binary and spike-and-slab latent factors. We introduce a non-negative constraint in the spike-and-slab prior that eliminates the usual sign ambiguity present in factor models and lowers the generaliza-tion error on the datasets tested here. For the generative models we use probit functions, which can be sampled without tuning param-eters, unlike previous works that used logistic functions. The posterior distributions involve mixtures of binary and truncated Gaussian variables, for which we present exact Hamil-tonian Monte Carlo samplers and compare their properties to Gibbs samplers. 1