Gerhard Widmer. Recognition and exploitation of contextual clues via incremental meta-learning. In Proceedings of the Thirteenth International Conference on Machine Learning, pages 525--533. Morgan Kaufmann, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... prediction (e.g. Dietterich and Michalski s work with Eleusis [DM86] temporal logics and their applications to recognising events, for example Kumar s work on temporal event conceptualisation [KM87] and the recent work in context detection and extraction for machine learning applications [Wid96, HHS98] While some of these areas bear interesting relations to temporal classification, they di#er in several regards. Sequence prediction is not about learning from labelled examples. The event conceptualisation work focuses on recognition of temporal events, but not learning the events ....

Gerhard Widmer. Recognition and exploitation of contextual clues via incremental meta-learning. In Proceedings of the Thirteenth International Conference on Machine Learning, pages 525--533. Morgan Kaufmann, 1996.

....Instance based learning could be used as the Splice 2R underlying learner, any hidden contexts thereby recognised by Splice 2R could then be utilised with these techniques for on line prediction. A somewhat different on line method designed to detect and exploit contextual attributes is MetaL(B) [31]. In this case, contextual attributes are considered to be predictive of the relevance of other attributes. MetaL(B) works by using the detected contextual attributes to trigger changes to the set of features presented to the classifier. While this approach and context definition is quite ....

Gerhard Widmer. Recognition and exploitation of contextual clues via incremental meta-learning. In Lorenza Saitta, editor, Machine Learning: Proceedings of the Thirteenth International Workshop, pages 525-- 533, San Francisco, 1996. Morgan Kaufmann.

....they cannot be applied directly to solve real world problems. Many challenges can only be found in real world applications. The changing environment can cause the data to fluctuate and make the previously discovered patterns partially invalid. Such phenomenon is termed as concept drift (Widmer, 1996). The possible solutions include incremental methods for updating the patterns and treating such drifts as an opportunity for interesting discovery by using it to cue the search for patterns of change (Matheus et al. 1994) In practice, we often find many organizations collecting data as a ....

Widmer, G.: 1996, `Recognition and Exploitation of Contextual Clues via Incremental Meta-Learning'. In: L. Saitta (ed.): Machine Learning: Proceedings of the Thirteenth International Conference. Bari, Italy, pp. 525--533.

....2.6 Context and machine learning The meaning of many concepts heavily depends on some implicit context, and changes in that context can cause more or less radical changes in the concepts. Incremental concept learning in such domains requires the ability to recognize and adapt to such changes. Widmer (1996) presents a general two level learning model, and its realization in his METAL(B) system. This system can learn to detect certain types of contextual clues, and can react accordingly with context changes. The model consists of a base level learner that perfoms the regular online learning and ....

Widmer G. (1996) "Recognition and exploitation of contextual clues via incremental metalearning (Extended version)", e-mail to "gerhard@ai.univie.ac.at".

....13 4 Expanding Features In the above two sections, we assume that the features describing data are fixed once and for all. It is only a matter of which features are irrelevant. In some real world tasks, however, the concepts of interest may not be entirely stable they change over time (Widmer, 1996). One way of viewing concept drifts is to assume that the concepts of interest depend on some context, and that changes in this context include corresponding changes in the target concepts. One type of concept drifts may be captured by changing values of features. In some domains, systematic ....

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental metalearning. In L. Saitta (Ed.), Machine Learning: Proceedings of the Thirteenth International Conference (p. 525-533). Bari, Italy: Morgan Kaufmann Publishers.

....a very simple heuristic was used to select the relevant local concept employed for prediction. More complex domains will require more complex methods to identify the relevant local concept. One approach may be to reason about duration and likely sequences of the identified contexts as suggested in (Widmer, 1996). 5 CONCLUSION A new meta learning algorithm, Splice, enables an underlying machine learning system to learn local concepts from domains with hidden changes in context. We demonstrated that the Splice approach is viable in at least a simple domain and is robust in the presence of noise in the ....

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental meta-learning.

....a very simple heuristic was used to select the relevant local concept employed for prediction. More complex domains will require more complex methods to identify the relevant local concept. One approach may be to reason about duration and likely sequences of the identified contexts as suggested in (Widmer, 1996). 5 CONCLUSION A new meta learning algorithm, Splice, enables an underlying machine learning system to learn local concepts from domains with hidden changes in context. We demonstrated that Splice is viable in at least a simple domain and is robust in the presence of noise in the domain ....

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental meta-learning.

.... information neighbouring phonemes Watrous (1991) speech recognition sound spectrum information speaker s identity Watrous (1993) heart disease diagnosis electrocardiogram data patient s identity Watrous (1995) tonal music harmonization meter, tactus, local key to be discovered by the learner Widmer (1996) malization, contextual expansion, contextual classifier selection, and contextual classification adjustment. Strategy 1: Contextual normalization: Contextual features can be used to normalize context sensitive primary features, prior to classification. The intent is to process context sensitive ....

.... 1993b) Normalization, Expansion, Weighting Explicit Turney and Halasz (1993) Normalization Explicit Watrous (1991) Adjustment Explicit Watrous (1993) Normalization Explicit Watrous and Towell (1995) Adjustment Explicit Widmer and Kubat (1992, 1993, 1996) Selection Implicit temporal sequence Widmer (1996) Selection Explicit 6 Conclusion This paper briefly surveyed the literature on machine learning in context sensitive domains. We found that there are five basic strategies for managing context sensitive features and two strategies for recovering lost context. Combining strategies appears to be ....

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental meta-learning. Machine Learning: Proceedings of the 13th International Conference, California: Morgan Kaufmann.

....Instance based learning could be used as the Splice 2R underlying learner, any hidden contexts thereby recognised by Splice 2R could then be utilised with these techniques for on line prediction. A somewhat different on line method designed to detect and exploit contextual attributes is MetaL(B) [31]. In this case, contextual attributes are considered to be predictive of the relevance of other attributes. MetaL(B) works by using the detected contextual attributes to trigger changes to the set of features presented to the classifier. While this approach and context definition is quite ....

Gerhard Widmer. Recognition and exploitation of contextual clues via incremental meta-learning. In Lorenza Saitta, editor, Machine Learning: Proceedings of the Thirteenth International Workshop, pages 525-- 533, San Francisco, 1996. Morgan Kaufmann.

.... (e.g. Dietterich and Michalski s work with Eleusis [DM86] work with temporal logics and their applications to recognising events, for example Kumar s work on temporal event conceptualisation [KM87] and the recent work in context detection and extraction for machine learning applications [Wid96, HHS98] While some of these areas bear some interesting relations to temporal classification, they differ in several regards. Sequence prediction is not about learning from labelled examples, nor, typically does it deal with multivariate data. The event conceptualisation work focuses on ....

Gerhard Widmer. Recognition and exploitation of contextual clues via incremental meta-learning. In Proceedings of the Thirteenth International Conference on Machine Learning, pages 525--533. Morgan Kaufmann, 1996.

....a very simple heuristic was used to select the relevant local concept employed for prediction. More complex domains will require more complex methods to identify the relevant local concept. One approach may be to reason about duration and likely sequences of the identified contexts as suggested in (Widmer, 1996). Some on line learners can use the proceeding concept to seed the development of new concepts. Splice is unable to share structure across contexts. It is possible that this limitation will reduce the effectiveness of Splice in situations with slow drift. On the other hand, this characteristic is ....

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental meta-learning.

....of older instances. Stagger (Schlimmer and Granger, 1996) was the first reported machine learning system that dealt with hidden changes in context. This system dealt with changes in context by discarding any concepts that fell below a threshold accuracy. Splice is most related to the Flora (Widmer Kubat, 1996) family of on line learners. These adapt to hidden changes in context by updating the current concept to match a window of recent instances. Rapid adaptation to changes in context is assured by altering the window size in response to shifts in prediction accuracy and concept complexity. One ....

....On line classification is achieved by switching between stable concepts according to the current context. 3.2.1. STAGGER data set The data sets in the following experiments are based on those used to evaluate Stagger (Schlimmer Granger, 1986) and subsequently used to evaluate Flora (Widmer Kubat, 1996). While our approach is substantially different, use of the same data set al..lows some comparison of results. A program was used to generate data sets. This allows us to control the recurrence of contexts and other factors such as noise 1 and duration of contexts. The task has four attributes, ....

[Article contains additional citation context not shown here]

Widmer, G. (1996). Recognition and exploitation of contextual clues via incremental metalearning.

.... of context dependent concepts (see also Bergadano et al. 1992) Recently, context dependence has been recognized as a problem in a number of practical machine learning projects (e.g. Katz et al. 1990; Turney, 1993; Turney Halasz, 1993; Watrous, 1993; Watrous Towell, 1995; see also Kubat Widmer, 1996). There, various techniques for context normalization etc. were developed. All of these methods either assume that contextual attributes are explicitly identified by the user, or require separate pre training phases on special data sets that are cleanly separated according to context. We are ....

....Classifier with Meta Learning: MetaL(B) Our first algorithm is called MetaL(B) MetaLearning with underlying Bayes classifier) and uses a simple Bayesian classifier as its underlying ( object level ) incremental learning algorithm. It was first presented and is discussed in more detail in (Widmer, 1996). In MetaL(B) the contextual attributes identified by meta learning are used to focus the base level Bayesian classifier on relevant examples when making predictions: whenever a new instance comes in, the set of attributes that are currently contextual (if any) is established, and the Bayesian ....

[Article contains additional citation context not shown here]

Widmer, G. (1996). Recognition and Exploitation of Contextual Clues via Incremental MetaLearning.

No context found.

Widmer, G. (1996). Recognition and Exploitation of Contextual Clues via Incremental Meta-Learning. Proceedings of the 13th International Conference on Machine Learning, Bari, Italy

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