Fully-Automatic Bayesian Piecewise Sparse Linear Models

Riki Eto, Ryohei Fujimaki, Satoshi Morinaga, Hiroshi Tamano
Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics, PMLR 33:238-246, 2014.

Abstract

Piecewise linear models (PLMs) have been widely used in many enterprise machine learning problems, which assign linear experts to individual partitions on feature spaces and express whole models as patches of local experts. This paper addresses simultaneous model selection issues of PLMs; partition structure determination and feature selection of individual experts. Our contributions are mainly three-fold. First, we extend factorized asymptotic Bayesian (FAB) inference for hierarchical mixtures of experts (probabilistic PLMs). FAB inference offers penalty terms w.r.t. partition and expert complexities, and enable us to resolve the model selection issue. Second, we propose posterior optimization which significantly improves predictive accuracy. Roughly speaking, our new posterior optimization mitigates accuracy degradation due to a gap between marginal log-likelihood maximization and predictive accuracy. Third, we present an application of energy demand forecasting as well as benchmark comparisons. The experiments show our capability of acquiring compact and highly-accurate models.

Cite this Paper


BibTeX
@InProceedings{pmlr-v33-eto14, title = {{Fully-Automatic Bayesian Piecewise Sparse Linear Models}}, author = {Eto, Riki and Fujimaki, Ryohei and Morinaga, Satoshi and Tamano, Hiroshi}, booktitle = {Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics}, pages = {238--246}, year = {2014}, editor = {Kaski, Samuel and Corander, Jukka}, volume = {33}, series = {Proceedings of Machine Learning Research}, address = {Reykjavik, Iceland}, month = {22--25 Apr}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v33/eto14.pdf}, url = {https://proceedings.mlr.press/v33/eto14.html}, abstract = {Piecewise linear models (PLMs) have been widely used in many enterprise machine learning problems, which assign linear experts to individual partitions on feature spaces and express whole models as patches of local experts. This paper addresses simultaneous model selection issues of PLMs; partition structure determination and feature selection of individual experts. Our contributions are mainly three-fold. First, we extend factorized asymptotic Bayesian (FAB) inference for hierarchical mixtures of experts (probabilistic PLMs). FAB inference offers penalty terms w.r.t. partition and expert complexities, and enable us to resolve the model selection issue. Second, we propose posterior optimization which significantly improves predictive accuracy. Roughly speaking, our new posterior optimization mitigates accuracy degradation due to a gap between marginal log-likelihood maximization and predictive accuracy. Third, we present an application of energy demand forecasting as well as benchmark comparisons. The experiments show our capability of acquiring compact and highly-accurate models.} }
Endnote
%0 Conference Paper %T Fully-Automatic Bayesian Piecewise Sparse Linear Models %A Riki Eto %A Ryohei Fujimaki %A Satoshi Morinaga %A Hiroshi Tamano %B Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2014 %E Samuel Kaski %E Jukka Corander %F pmlr-v33-eto14 %I PMLR %P 238--246 %U https://proceedings.mlr.press/v33/eto14.html %V 33 %X Piecewise linear models (PLMs) have been widely used in many enterprise machine learning problems, which assign linear experts to individual partitions on feature spaces and express whole models as patches of local experts. This paper addresses simultaneous model selection issues of PLMs; partition structure determination and feature selection of individual experts. Our contributions are mainly three-fold. First, we extend factorized asymptotic Bayesian (FAB) inference for hierarchical mixtures of experts (probabilistic PLMs). FAB inference offers penalty terms w.r.t. partition and expert complexities, and enable us to resolve the model selection issue. Second, we propose posterior optimization which significantly improves predictive accuracy. Roughly speaking, our new posterior optimization mitigates accuracy degradation due to a gap between marginal log-likelihood maximization and predictive accuracy. Third, we present an application of energy demand forecasting as well as benchmark comparisons. The experiments show our capability of acquiring compact and highly-accurate models.
RIS
TY - CPAPER TI - Fully-Automatic Bayesian Piecewise Sparse Linear Models AU - Riki Eto AU - Ryohei Fujimaki AU - Satoshi Morinaga AU - Hiroshi Tamano BT - Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics DA - 2014/04/02 ED - Samuel Kaski ED - Jukka Corander ID - pmlr-v33-eto14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 33 SP - 238 EP - 246 L1 - http://proceedings.mlr.press/v33/eto14.pdf UR - https://proceedings.mlr.press/v33/eto14.html AB - Piecewise linear models (PLMs) have been widely used in many enterprise machine learning problems, which assign linear experts to individual partitions on feature spaces and express whole models as patches of local experts. This paper addresses simultaneous model selection issues of PLMs; partition structure determination and feature selection of individual experts. Our contributions are mainly three-fold. First, we extend factorized asymptotic Bayesian (FAB) inference for hierarchical mixtures of experts (probabilistic PLMs). FAB inference offers penalty terms w.r.t. partition and expert complexities, and enable us to resolve the model selection issue. Second, we propose posterior optimization which significantly improves predictive accuracy. Roughly speaking, our new posterior optimization mitigates accuracy degradation due to a gap between marginal log-likelihood maximization and predictive accuracy. Third, we present an application of energy demand forecasting as well as benchmark comparisons. The experiments show our capability of acquiring compact and highly-accurate models. ER -
APA
Eto, R., Fujimaki, R., Morinaga, S. & Tamano, H.. (2014). Fully-Automatic Bayesian Piecewise Sparse Linear Models. Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 33:238-246 Available from https://proceedings.mlr.press/v33/eto14.html.

Related Material