## Variational Methods for Inference and Estimation in Graphical Models (1997)

Citations: | 58 - 3 self |

### BibTeX

@TECHREPORT{Jaakkola97variationalmethods,

author = {Tommi S. Jaakkola},

title = {Variational Methods for Inference and Estimation in Graphical Models},

institution = {},

year = {1997}

}

### Years of Citing Articles

### OpenURL

### Abstract

Graphical models enhance the representational power of probability models through qualitative characterization of their properties. This also leads to greater efficiency in terms of the computational algorithms that empower such representations. The increasing complexity of these models, however, quickly renders exact probabilistic calculations infeasible. We propose a principled framework for approximating graphical models based on variational methods. We develop variational techniques from the perspective that unifies and expands their applicability to graphical models. These methods allow the (recursive) computation of upper and lower bounds on the quantities of interest. Such bounds yield considerably more information than mere approximations and provide an inherent error metric for tailoring the approximations individually to the cases considered. These desirable properties, concomitant to the variational methods, are unlikely to arise as a result of other deterministic or stochas...

### Citations

954 |
An introduction to Bayesian networks
- Jensen
- 1996
(Show Context)
Citation Context ...he efficiency of computational operations on them. (For readers who will find the remaining introduction hard to follow in terms of its content on graphical models, we suggest an introductory book by =-=Jensen, 1996-=- as a suitable background reading.) In graphical representation of probabilities the nodes in the graph correspond to the variables in the probability model and the edges connecting the nodes signify ... |

602 |
The computational complexity probabilistic inference using Bayesian belief networks
- Cooper
(Show Context)
Citation Context ...en comes with a sizable computational cost of evaluating the conclusion implied by the probabilistic framework. Probabilistic inference, after all, has been shown to be an NP-hard problem in general (=-=Cooper 1990-=-). The size and complexity of the QMR belief network in particular appears to render exact evaluation methods inapplicable. To reap the benefits from the probabilistic formulation of the QMR network, ... |

586 | Probabilistic inference using Markov chain Monte Carlo methods
- Neal
- 1993
(Show Context)
Citation Context ... regression---and show how variational approximation techniques can restore the computational feasibility of the Bayesian formalism. Variational methods should be contrasted with sampling techniques (=-=Neal 1993-=-) that have become standard in the context of Bayesian calculations. While surely powerful in evaluating complicated integrals, sampling techniques do not guarantee monotonically improving approximati... |

184 | Large Deviations Techniques in Decision, Simulation, and Estimation - Bucklew - 1990 |

29 |
An empirical analysis of likelihoodâ€”Weighting simulation on a large, multiply connected medical belief network
- Shwe, Cooper
- 1991
(Show Context)
Citation Context ...c model. Second, we should be able to assess the reliability of the results when such computational resources are not available. In the context of the QMR-belief network, Shwe et al. (1991) (see also =-=Shwe & Cooper 1991-=-) proposed the use of sampling techniques to obtain estimates of the posterior probabilities for the diseases (diagnoses in probabilistic terms). Sampling methods meet the first of our objectives sinc... |

4 |
Symbolic probabilistic inference in large BN20 networks
- D'Ambrosio
- 1994
(Show Context)
Citation Context ... the vectors of positive and negative findings. The computation of these posterior probabilities exactly is in the worst case exponentially costly in the number of positive findings (Heckerman, 1988; =-=D'Ambrosio 1994-=-); the negative findings f \Gamma , on the other hand, can be incorporated in linear time (in the number of associated diseases and in the number of negative findings). In practical diagnostic situati... |

4 | Probabilistic diagnosis using a reformulation of the INTERNIST-1/QMR knowledge base, I â€“ the probabilistic model and inference algorithms - Cooper - 1991 |