Zero-Truncated Poisson Model for Scalable Bayesian Factorization of Massive Binary Tensors with Mode-Networks

Changwei Hu Duke University, Piyush Rai Duke University, Lawrence Carin Duke University
Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence, PMLR R13:950-959, 2015.

Abstract

We present a scalable Bayesian model for low-rank factorization of massive tensors with binary observations. The proposed model has the following key properties: (1) in contrast to the models based on logistic or probit likelihood, using a zero-truncated Poisson likelihood for binary data allows our model to scale up in the number of ones in the tensor, without sacrificing on the quality of the results; (2) side-information in form of binary pairwise relationships (e.g., an adjacency network) between objects in any tensor mode can also be leveraged, which can be especially useful in “cold-start” settings; and (3) the model admits simple inference via batch, as well as online MCMC; the latter allows us to scale up even for dense binary data (i.e., when the number of ones in the tensor/network is also massive). In addition, non-negative factor matrices in our model provide easy interpretability, and the tensor rank is inferred from data. We apply our model on several real-world massive binary tensors, and on massive binary tensors with binary mode-network(s) as side-information.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR13-university15v, title = {Zero-Truncated {P}oisson Model for Scalable {B}ayesian Factorization of Massive Binary Tensors with Mode-Networks}, author = {University, Changwei Hu Duke and University, Piyush Rai Duke and University, Lawrence Carin Duke}, booktitle = {Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence}, pages = {950--959}, year = {2015}, editor = {Meila, Marina and Heskes, Tom}, volume = {R13}, series = {Proceedings of Machine Learning Research}, month = {12--16 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r13/main/assets/university15v/university15v.pdf}, url = {https://proceedings.mlr.press/r13/university15v.html}, abstract = {We present a scalable Bayesian model for low-rank factorization of massive tensors with binary observations. The proposed model has the following key properties: (1) in contrast to the models based on logistic or probit likelihood, using a zero-truncated Poisson likelihood for binary data allows our model to scale up in the number of ones in the tensor, without sacrificing on the quality of the results; (2) side-information in form of binary pairwise relationships (e.g., an adjacency network) between objects in any tensor mode can also be leveraged, which can be especially useful in “cold-start” settings; and (3) the model admits simple inference via batch, as well as online MCMC; the latter allows us to scale up even for dense binary data (i.e., when the number of ones in the tensor/network is also massive). In addition, non-negative factor matrices in our model provide easy interpretability, and the tensor rank is inferred from data. We apply our model on several real-world massive binary tensors, and on massive binary tensors with binary mode-network(s) as side-information.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Zero-Truncated Poisson Model for Scalable Bayesian Factorization of Massive Binary Tensors with Mode-Networks %A Changwei Hu Duke University %A Piyush Rai Duke University %A Lawrence Carin Duke University %B Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2015 %E Marina Meila %E Tom Heskes %F pmlr-vR13-university15v %I PMLR %P 950--959 %U https://proceedings.mlr.press/r13/university15v.html %V R13 %X We present a scalable Bayesian model for low-rank factorization of massive tensors with binary observations. The proposed model has the following key properties: (1) in contrast to the models based on logistic or probit likelihood, using a zero-truncated Poisson likelihood for binary data allows our model to scale up in the number of ones in the tensor, without sacrificing on the quality of the results; (2) side-information in form of binary pairwise relationships (e.g., an adjacency network) between objects in any tensor mode can also be leveraged, which can be especially useful in “cold-start” settings; and (3) the model admits simple inference via batch, as well as online MCMC; the latter allows us to scale up even for dense binary data (i.e., when the number of ones in the tensor/network is also massive). In addition, non-negative factor matrices in our model provide easy interpretability, and the tensor rank is inferred from data. We apply our model on several real-world massive binary tensors, and on massive binary tensors with binary mode-network(s) as side-information. %Z Reissued by PMLR on 04 October 2026.
APA
University, C.H.D., University, P.R.D. & University, L.C.D.. (2015). Zero-Truncated Poisson Model for Scalable Bayesian Factorization of Massive Binary Tensors with Mode-Networks. Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R13:950-959 Available from https://proceedings.mlr.press/r13/university15v.html. Reissued by PMLR on 04 October 2026.

Related Material