M. Harries and K. Horn. Learning stable concepts in domains with hidden changes in context. In 13th ICML, workshop on Learning in Context Sensitive Domains, Bari, Italy, July 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....context sensitive Bayesian classifiers that switch Bayesian classifiers corresponding to different contexts. We train each component classifier by an efficient training algorithm, expectation propagation [6] Similar to classical context learning algorithms, for example, the Splice algorithm [2], the context sensitive Bayesian classifier can be viewed as an approximation to a mixture of experts [4, 9] with an easier training procedure. Our algorithm is also an online learning technique: after training on current data sets, if there are more data sets possibly containing different ....

M. Harries and K. Horn. Learning stable concepts in domains with hidden changes in context. In 13th ICML, workshop on Learning in Context Sensitive Domains, Bari, Italy, July 1996.

....clusters so, that some parameters of the classification process (context or classifiers responsibility) can be considered to be stable within each cluster. A clustering algorithm, called contextual clustering, was successfully applied in domains with hidden changes in the context of the concepts [7]. The whole domain was partitioned into clusters according to the apparent similarity of their contexts. The context was supposed to be stable within each cluster. The clustered algorithm based on this batch learner had accuracy up to 30 better, than a non clustered batch learner did. This paper ....

Harries, M., Horn, K.: Learning stable concepts in domains with hidden changes in context. In: Proceedings of the Thirteenth International Conference on Machine Learning. Workshop on Learning in Context Sensitive Domains, Bari, Italy, July 3-6 (1996)

....set of identified context changes can be refined by contextual clustering. Contextual clustering combines similar intervals of the dataset, where the similarity of two intervals is based upon the degree to which a partial model is accurate on both intervals. 2 SPLICE 1 6 2 Splice 1 Splice 1 [8] is designed to recognise stable context and extract local concepts from domains with hidden changes in context. Splice 1 can use any ordinal attribute as the environmental attribute, in order to preserve clarity in the following discussion we have substituted Time for the broader term ....

M. Harries and K. Horn. Learning stable concepts in domains with hidden changes in context. In M. Kubat and G Widmer, editors, Learning in context-sensitive domains (Workshop Notes). 13th International Conference on Machine Learning, Bari, Italy, 1996.

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