Neural Variational Inference and Learning in Belief Networks

Andriy Mnih, Karol Gregor
; Proceedings of the 31st International Conference on Machine Learning, PMLR 32(2):1791-1799, 2014.

Abstract

Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference network gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset.

Cite this Paper


BibTeX
@InProceedings{pmlr-v32-mnih14, title = {Neural Variational Inference and Learning in Belief Networks}, author = {Andriy Mnih and Karol Gregor}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {1791--1799}, year = {2014}, editor = {Eric P. Xing and Tony Jebara}, volume = {32}, number = {2}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/mnih14.pdf}, url = {http://proceedings.mlr.press/v32/mnih14.html}, abstract = {Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference network gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset.} }
Endnote
%0 Conference Paper %T Neural Variational Inference and Learning in Belief Networks %A Andriy Mnih %A Karol Gregor %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-mnih14 %I PMLR %J Proceedings of Machine Learning Research %P 1791--1799 %U http://proceedings.mlr.press %V 32 %N 2 %W PMLR %X Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference network gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset.
RIS
TY - CPAPER TI - Neural Variational Inference and Learning in Belief Networks AU - Andriy Mnih AU - Karol Gregor BT - Proceedings of the 31st International Conference on Machine Learning PY - 2014/01/27 DA - 2014/01/27 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-mnih14 PB - PMLR SP - 1791 DP - PMLR EP - 1799 L1 - http://proceedings.mlr.press/v32/mnih14.pdf UR - http://proceedings.mlr.press/v32/mnih14.html AB - Highly expressive directed latent variable models, such as sigmoid belief networks, are difficult to train on large datasets because exact inference in them is intractable and none of the approximate inference methods that have been applied to them scale well. We propose a fast non-iterative approximate inference method that uses a feedforward network to implement efficient exact sampling from the variational posterior. The model and this inference network are trained jointly by maximizing a variational lower bound on the log-likelihood. Although the naive estimator of the inference network gradient is too high-variance to be useful, we make it practical by applying several straightforward model-independent variance reduction techniques. Applying our approach to training sigmoid belief networks and deep autoregressive networks, we show that it outperforms the wake-sleep algorithm on MNIST and achieves state-of-the-art results on the Reuters RCV1 document dataset. ER -
APA
Mnih, A. & Gregor, K.. (2014). Neural Variational Inference and Learning in Belief Networks. Proceedings of the 31st International Conference on Machine Learning, in PMLR 32(2):1791-1799

Related Material