Adaptive probabilistic networks with hidden variables (1997)by J. Binder, D. Koller, S.J. Russell, and K. Kanazawa
Abstract:
Probabilistic networks (also known as Bayesian belief networks) allow a compact description of complex stochastic relationships among several random variables. They are rapidly becoming the tool of choice for uncertain reasoning in artificial intelligence. In this paper, we investigate the problem of learning probabilistic networks with known structure and hidden variables. This is an important problem, because structure is much easier to elicit from experts than numbers, and the world is rarely fully observable. We present a gradientdescent based algorithm, and show that the gradient can be computed locally, using information that is available as a byproduct of standard probabilistic network inference algorithms. Our results demonstrate that using prior knowledge about the structure, even with hidden variables, can significantly improve the learning rate of probabilistic networks. We extend the method to networks where the conditional probability tables are described using a small number of parameters. Examples include noisyOR nodes and dynamic probabilistic networks. We show how this additional structure can be exploited by our algorithm to speed up the learning even further. We also suggest how similar ideas can be used to learn hybrid networks, where some of the nodes take on values in a continuous domain.
Download Information
J. Binder, D. Koller, S.J. Russell, and K. Kanazawa (1997). "Adaptive probabilistic networks with hidden variables." Machine Learning, 29(23), 213244.
Full version of IJCAI '95 paper.


Bibtex citation
@article{Binder+al:MLJ97,
author = "J. Binder and D. Koller and S.J. Russell and K. Kanazawa",
title = "Adaptive probabilistic networks with hidden
variables",
journal = "Machine Learning",
volume = "29",
number = "23",
pages = "213244",
year = "1997",
note = {Full version of IJCAI '95 paper},
}
full list
