Comparative Study of Inference Methods for Bayesian Nonnegative Matrix\n Factorisation

2017/07/13 by Thomas Brouwer, Jes Frellsen, Brouwer, Thomas +3
Biochemistry, Genetics and Molecular Biology · Computer Science · #Bayesian Methods and Mixture Models #FOS: Computer and information sciences #Gene expression and cancer classification #Genetic and phenotypic traits in livestock #Machine Learning (cs.LG) #Machine Learning (stat.ML)

paper · pdf · doi:10.48550/arxiv.1707.05147

openalex publication_date 2017/07/13 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

In this paper, we study the trade-offs of different inference approaches for\nBayesian matrix factorisation methods, which are commonly used for predicting\nmissing values, and for finding patterns in the data. In particular, we\nconsider Bayesian nonnegative variants of matrix factorisation and\ntri-factorisation, and compare non-probabilistic inference, Gibbs sampling,\nvariational Bayesian inference, and a maximum-a-posteriori approach. The\nvariational approach is new for the Bayesian nonnegative models. We compare\ntheir convergence, and robustness to noise and sparsity of the data, on both\nsynthetic and real-world datasets. Furthermore, we extend the models with the\nBayesian automatic relevance determination prior, allowing the models to\nperform automatic model selection, and demonstrate its efficiency.\n

Related

• Flexible and Hierarchical Prior for Bayesian Nonnegative Matrix Factorization
• Prior and Likelihood Choices for Bayesian Matrix Factorisation on Small Datasets
• A particle-based variational approach to Bayesian Non-negative Matrix Factorization
• Fast Bayesian Non-Negative Matrix Factorisation and Tri-Factorisation
• Bayesian Mean-parameterized Nonnegative Binary Matrix Factorization