N. Friedman and M. Goldszmidt. Building classi ers using bayesian networks. In AAAI/IAAI, Vol. 2, pages 1277-1284, 1996.  @ NUS  Home/Search   Document Not in Database   Summary   Related Articles   Check

....located between these two perspectives; it eliminates the disadvantages of both while exploiting their advantages. The rst induction of classi ers involves a search over all possible networks and has been successfully solved in [2] and [4] However, it can be considered as unsupervised learning [7] since it does not distinguish between attribute variables and the class variable. Thus the results for a classi cation task are not suciently accurate. The second induction approach is based on the re nement of the naive Bayes classi er, which has already proved its power for classi cation in ....

....Bayes classi er, which has already proved its power for classi cation in many applications [13] Due to the fact, that this classi er comes with the strong assumption of independence, re nements are achieved by relaxing this assumption. Signi cant work in that eld is found in [16] 14] and [7]. The latter approach improves the naive Bayes classi er by capturing single dependencies between the attributes. Our approach is motivated by this one but extends it by two new features. These two features allow the possibility of learning multiple correlations between attributes and the ....

[Article contains additional citation context not shown here]

N. Friedman and M. Goldszmidt. Building Classi ers using Bayesian Networks. In Thirteenth National Conf. on Arti cial Intelligence, 1996.

....develop methods for learning the simplest BNs that t the data adequately. Some examples of this tradeo between the cost of the learning process and the quality of the learnt models are NB models [14, 43] Extended Naive Bayes (ENB) models [36, 37, 39, 42] and Tree Augmented Naive Bayes models [18, 19, 20, 26, 32, 40]. Despite the wide recognition that these models are a weaker representation of some domains than more general BNs, the expressive power of these models is often acceptable. Moreover, these models appeal to human intuition and can be learnt relatively quickly. For the sake of brevity, the class ....

Friedman, N., & Goldszmidt, M. (1996). Building Classiers using Bayesian Networks. Proccedings of the Thirteenth National Conference on Articial Intelligence (pp. 1277-1284). Menlo Park, CA: AAAI Press.

....in Section 3.2.2 to automatically distinguish between relevant and irrelevant features for learning. As we have addressed, we use the BS EM algorithm as our unsupervised learning algorithm. In the current experiments, we limit the BS EM algorithm to learn Tree Augmented Naive Bayes (TANB) models [14, 30, 36]. This is a sensible and usual decision to reduce the otherwise large search space of CGNs. Moreover, this allows to solve eciently data clustering problems of considerable size as it is well known the diculty involved in learning densely connected CGNs from large databases, and the painfully slow ....

N. Friedman and M. Goldszmidt, \Building Classiers Using Bayesian Networks," Proceedings of the Thirteenth National Conference on Articial Intelligence, AAAI Press, Menlo Park, CA, pp. 1277-1284, 1996.

....to develop methods for learning simple conditional Gaussian networks while preserving the quality of the learnt models. A symmetrical problem is found when working in discrete domains, and some of the proposed solutions are based on achieving such a balance between eciency and e ectiveness [7, 10, 11, 12, 19, 24, 27, 28, 29, 30, 31, 32, 33, 34]. Keeping this idea in mind, we propose two classes of compromise conditional Gaussian networks for data clustering. The models of the rst class belong to what in [10, 11, 12, 19] are called tree augmented naive Bayes models (more recently referred to 2 as mixtures of trees with shared ....

....are based on achieving such a balance between eciency and e ectiveness [7, 10, 11, 12, 19, 24, 27, 28, 29, 30, 31, 32, 33, 34] Keeping this idea in mind, we propose two classes of compromise conditional Gaussian networks for data clustering. The models of the rst class belong to what in [10, 11, 12, 19] are called tree augmented naive Bayes models (more recently referred to 2 as mixtures of trees with shared structure [24] These models are used in the referred works as Bayesian classi ers, while in [31, 33] they are evaluated in the framework of data clustering for discrete domains. In this ....

[Article contains additional citation context not shown here]

Friedman, N. and Goldszmidt, M., Building Classiers using Bayesian Networks, Proceedings of the Thirteenth National Conference on Articial Intelligence, AAAI Press, Menlo Park, CA, 1277-1284, 1996.

....of survival after one year of being diagnosed. The accuracy of the obtained models for the predictions of survival after one, three and ve years of being diagnosed is, respectively, of 94.4 , 80.4 and 72.0 . 3.1.2 Tree Augmented Naive Bayes. TAN GA Several authors see, for instance [8] have tried to construct models that starting from a Bayesian Network structure in which all variables are conditionally independent given the variable to predict, successively add arcs to the structure in a greedy way. These arcs will re ect correlations among variables and its inclusion ....

Friedman, N., and Goldszmidt, M. (1996) Building Classiers using Bayesian Networks. Proceedings of AAAI-96.

....with one of the top classi ers, Na ve Bayes (Friedman, Geiger and Goldszmidt 1997) 4. 1 Validation We compared the performance of our approach with a Na ve Bayes implementation from Carnegie Mellon University (McCallum 1998) which has been found to be among the most e ective classi ers (Friedman and Goldszmidt 1996). In the Na ve Bayes approach (Langley, Iba and Thompson 1992) the conditional probability of each attribute value given a particular category is determined as well as the probability of the category appearing. This probabilistic information is typically established through a learning procedure, ....

Friedman, N. and M. Goldszmidt (1996). Building classiers using Bayesian networks. In Thirteenth National Conference on Articial Intelligence (AAAI).

No context found.

N. Friedman and M. Goldszmidt. Building classi ers using bayesian networks. In AAAI/IAAI, Vol. 2, pages 1277-1284, 1996.

No context found.

N. Friedman and M. Goldszmidt. Building classi ers using bayesian networks. In Proceedings of National Conferrence on Arti cial Intelligence (AAAI-96), pages 1277-1284, 1996.

No context found.

N. Friedman and M. Goldszmidt. Building classi ers using bayesian networks. In AAAI/IAAI, Vol. 2, pages 1277-1284, 1996.

No context found.

N. Friedman and M. Goldszmidt, \Building Classi ers using Bayesian networks," Proceedings: 13th National Conference on Arti cial Intelligence, Portland, Oregon, pp. 1277-1284, 1996.

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

CiteSeer.IST at NUS - Copyright Penn State and NEC. Hosted by the School of Computing, National University of Singapore.