D. Heckerman and D. Chickering. A comparison of scientific and engineering criteria for bayesian model selection. In Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, pages 275--281, Ft. Lauderdale, Florida, January 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....components in the Gaussian mixture models in (1) as well as in (4) which amounts to the problem of model selection. The maximum likelihood approach is not appropriate for this task, as it would always prefer more components. Several techniques have been proposed under the topic of model selection [8] [11] Full Bayesian model selection techniques provide a more principled method of model selection and generally use a parametric or hierarchical form to approximate the prior distribution over the parameters [8] The BIC or MDL approach, as proposed by Rissanen [11] can be shown to be ....

....components. Several techniques have been proposed under the topic of model selection [8] 11] Full Bayesian model selection techniques provide a more principled method of model selection and generally use a parametric or hierarchical form to approximate the prior distribution over the parameters [8]. The BIC or MDL approach, as proposed by Rissanen [11] can be shown to be asymptotically consistent version of the full Bayesian model selection techniques. In the MDL technique, the description length (DL) is given as, DL = log p(X ) l 2) log n (9) where X is the data, is the parameter ....

D. Heckerman and D. Chickering, "A Comparison of Scientific and Engineering Criteria for Bayesian Model Selection," Technical Report MSR-TR-96-12, Microsoft Research, June, 1996.

....of R 2 s behavior generated by each of these models can be combined with weights equal to probabilities associated with the models to yield the 6 See [17] for more formal definition and more details. 6 overall prediction of R 2 s behavior. This is called Bayesian model averaging; see [17, 23] for further details. To handle the issue of predicting the other agent s action, while the other agent attempting to do the same, we suggest a knowledge based approach. Intuitively, instead of attempting to guess what the other agent will do, based on what its guess is as to what the original ....

David Heckerman and David Maxwell Chickering. A comparison of scientific and engineering criteria for bayesian model selection. Technical Report MSR-TR-96-12, Microsoft Research, Microsoft Corporation, Redmond, WA, 1996.

.... of the other agents resulting from a model M (R i ;ff) as p (R i ;ff) a 1 k : a n p , we can express the overall intentional probability of the other agents joint moves, p R i a 1 k : a n p , as an average over all possible models (this is known as Bayesian model averaging [38]) p R i a 1 k : a n p = X ff p R i ff Theta p (R i ;ff) a 1 k : a n p : 6) The joint probability, p (R i ;ff) a 1 k : a n p , of the other agents behaviors resulting from a model M (R i ;ff) can in turn be expressed as a product of the intentional ....

David Heckerman and David Maxwell Chickering. A comparison of scientific and engineering criteria for bayesian model selection. Technical Report MSR-TR-96-12, Microsoft Research, Microsoft Corporation, Redmond, WA, 1996.

....of other approximations based on such a coarse measure. The relationship between the stochastic complexity measure (called scientific criterion ) and crossvalidation measure (called engineering criterion ) together with some experimental results in model class selection tasks are discussed in [12]. In an earlier study [15] we have demonstrated that for some commonly used benchmark data sets, on the average very small random samples (less than 10 ) are sufficient to construct good predictive models. By good models we mean that they provide prediction performance comparable to the ....

D. Heckerman and D. Chickering. A comparison of scientific and engineering criteria for bayesian model selection. In Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, pages 275--281, Ft. Lauderdale, Florida, January 1997.

....3 This is not the only criteria one could imagine optimizing here. For example, one could be interested in finding a simple model of the underlying phenomenon that gives some insight into its fundamental nature, rather than simply producing a function that predicts well on future test examples (Heckerman Chickering 1996). However, we will focus on the traditional machine learning goal of minimizing prediction error. 4 One could consider more elaborate strategies that choose hypotheses from outside the sequence; e.g. by averaging several hypotheses together (Opitz Shavlik 1996; Breiman 1994) However, we will ....

Heckerman, D., and Chickering, D. 1996. A comparison of scientific and engineering criteria for Bayesian model selection. Technical Report MSR-TR-96-12, Microsoft Research.

No context found.

D. Heckerman and D. Chickering. A comparison of scientific and engineering criteria for bayesian model selection. In Proceedings of the Sixth International Workshop on Artificial Intelligence and Statistics, pages 275--281, Ft. Lauderdale, Florida, January 1997.

No context found.

David Heckerman and David Maxwell Chickering. A comparison of scientific and engineering criteria for Bayesian model selection. Technical Report MSR-TR-96-12, Microsoft Research, 1996.

No context found.

D. Heckerman, D. Chickering, A comparison of scientific and engineering criteria for Bayesian model selection, Technical Report MSR-TR-96-12, Microsoft Research, 1996.

No context found.

Heckerman, D.; and Chickering D.M., 1996. A comparison of scientific and engineering criteria for Bayesian model selection. Technical Report, MSR-TR-96-12, http://research.microsoft.com/ research/dtg/heckerma/TR-96-12.htm

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC