Bayesian Multi-View Tensor Factorization

This software package implements the first Bayesian Multi-View Tensor Factorization model, that finds the joint factorization of multiple coupled tensors. The code also provides the first Bayesian Tensor Canonical Correlation Analysis (CCA). The model assumes an underlying CNDECOMP/PARAFAC (CP) factorization and the results in our publication demonstrate that it can be also used as a well regularized Bayesian CP decompsition that can overcome degeneracy and find the true number of components even if there is missing data.

Source Code

If you use any of the algorithms implemented in the package, please cite the following paper. The zip-package contains both, the code for learning the model and an example script running on a toy dataset.

Download: BMTF.zip (R Implementation, Version 1.0, June 2014)
Dependencies: R package compiler
Example: The package contains a example script demo.R that runs on toy data.
License: Licensed under the FreeBSD open source license.

Publications

• Suleiman A. Khan, S Kaski, Bayesian Multi-view Tensor Factorization, European Conference on Machine Learning (ECML 2014), Nancy, France, 2014. PDF

Contact/Feedback

Suleiman Ali Khan: khan.suleiman@gmail.com
Author: Suleiman A. Khan