MACAU
Scalable Bayesian factorization with high-dimensional side information using MCMC
Description:
An MCMC-based factorization method Macau that is able to scale to millions of features, Fu > 1,000,000 while the factorized matrix Y can have a million of rows (and/or columns). In contrast, the sampling-based Bayesian factorization methods Bayesian Matrix Factorization with Side Information (BMFSI) and Bayesian Canonical PARAFAC with Features and Networks (BCPFN) have complexity O(Fu^3) because they require explicit computation and Cholesky decomposition of the covariance matrix of size Fu×Fu for linking the side information to the factorization.
More Information:
Originates from the AI-Driven Data Science research challenge (FAIR 2019–2023).
Responsible Research Lead: Yves Moreau
Authors: Jaak Simm, Adam Arany, Yves Moreau
Programming Language: Python/Cython/C++
Categories: Supervised learning, Recommender systems
Publications:
J. Simm, A. Arany, P. Zakeri, T. Haber, J. K. Wegner, V. Chupakhin, H. Ceulemans, Y. Moreau, "Macau: Scalable Bayesian factorization with high-dimensional side information using MCMC", In proceedings of the 2017 IEEE 27th International Workshop on Machine Learning for Signal Processing (MLSP), 2017