Kearns, M. and Seung, S. (1995). Learning from a Population of Hypotheses. Machine Learning, 18, 255-276  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... 1995; Maclin and Shavlik, 1995 ] In the present application, Sequel combined classifiers that were trained using different input features (see Section 6) The second aspect of creating an ensemble is the choice of the function for combining the predictions of the component classifiers [ Kearns and Seung, 1995 ] Examples of combination functions include voting schemes [ Hansen and Salamon, 1990 ] simple averages [ Lincoln and Skrzypek, 1989 ] weighted average schemes [ Perrone and Cooper, 1994; Rogova, 1994 ] and schemes for training combiners [ Rost and Sander, 1993; Wolpert, 1992; Zhang et ....

M. Kearns and H. Seung. Learning from a population of hypotheses. Machine Learning, 18:255--276, 1995.

....listed above. The latter two fall into the second category but aim to select hypotheses generated according to different concept description language and selection biases. Researchers in computational learning theory have also analysed aspects of the area of multiple learners. These include Kearns and Seung (1995) who model the problem using an oracle . Normally, each call to the oracle returns a single classified example. They show that to model a system with multiple learners, each call to the oracle should return a summary of the examples seen by a single 8 agent. Cesa Bianchi et al. 1995) also ....

M. Kearns and H. S. Seung (1995). Learning from a Population of Hypotheses, Machine Learning, 18:255-276.

....varying the set of input features, since the resulting component classification problems in fact differ significantly when the features are not completely redundant. The second aspect of creating an ensemble is the choice of the function for combining the predictions of the component classifiers (Kearns Seung 1995). Examples of combination functions include voting schemes (Hansen Salamon 1990) simple averages (Lincoln Skrzypek 1989) weighted average schemes (Perrone Cooper 1994; Rogova 1994) and schemes for training combiners (Rost Sander 1993; Wolpert 1992; Zhang, Mesirov, Waltz 1992) Clemen ....

Kearns, M., and Seung, H. 1995. Learning from a population of hypotheses. Machine Learning 18:255-- 276.

....activity in the area, a number of additional systems and methods have been developed, e.g. Silver et al. 1991; Svatek, 1994; Provost Hennessey, 1994; Chan Stolfo, 1995) Researchers in computational learning theory have also analysed aspects of the area of multiple learners. These include Kearns and Seung (1995) who model the problem using an oracle . Normally, each call to the oracle returns a single classified example. They show that to model a system with multiple learners, each call to the oracle should return a summary of the examples seen by a single agent. Cesa Bianchi et al. 1995) also address ....

M. Kearns and H. S. Seung (1995). Learning from a Population of Hypotheses, Machine Learning, 18:255-276.

....2 Combining Multiple Networks The basic idea in combining neural networks is to train a number of networks, and then somehow use the collection to increase generalization. Figure 2 shows a framework for combining multiple networks. The choice of a function for combining predictions is important [ Kearns and Seung, 1995 ] Examples of combination functions include voting schemes [ Hansen and Salamon, 1990 ] simple averages [ Lincoln and Skrzypek, 1990 ] weighted average schemes [ Perrone and Cooper, 1994; Rogova, 1994 ] and schemes for training combiners [ Rost and Sander, 1993; Wolpert, 1992; Zhang et ....

M. Kearns and H. Seung. Learning from a population of hypotheses. Machine Learning, 18:255--276, 1995.

....include (weighted) majority vote, weighted averaging, Bayesian likelihood combination and distribution summation. None of this work uses a learning algorithm to perform level 1 learning. Ting Low (1997) study the base line behavior of dagging empirically. Theoretical work on dagging includes Kearns Seung (1995) and Meir (1994) 7 Conclusions and future work This paper shows how stacked generalization can be successfully applied to combine bagged or dagged models derived from a single or multiple learning algorithms. Stacking using MLR almost always yields a lower predictive error rate than majority ....

Kearns, M. & H.S. Seung (1995), Learning from a Population of Hypotheses, Machine Learning, 18, pp. 255276, Kluwer Academic Publishers.

....a small difference in predictive error rate, then a model combination approach should be employed. Otherwise, use the model which derived using all the available data. Our empirical result seems to be stronger, in terms of the expected performance of model combination, than a theoretical result (Kearns Seung, 1995) based on the same working assumption. This theoretical work seeks . the possibility of somehow combining the independent hypotheses in a way that considerably outperforms any single hypothesis (Kearns Seung, 1995) The single hypothesis refers to any of the model combination s constituent ....

....of the expected performance of model combination, than a theoretical result (Kearns Seung, 1995) based on the same working assumption. This theoretical work seeks . the possibility of somehow combining the independent hypotheses in a way that considerably outperforms any single hypothesis (Kearns Seung, 1995). The single hypothesis refers to any of the model combination s constituent models. Our result indicates that model combination can significantly outperform not only its constituent models but the model learned from aggregating the available data; in spite of the the fact that this result only ....

Kearns, M. & H.S. Seung (1995), Learning from a Population of Hypotheses, Machine Learning, 18, pp. 255-276, Kluwer Academic Publishers.

....Stolfo, 1995. There have also been a number of workshops, as well as a comprehensive survey of learning in Distributed Artificial Intelligence (Weiss and Sen, 1995) Researchers in the field of computational learning theory have also analysed aspects of the area of multiple learners. These include Kearns and Seung (1995) who model inductive inference using an oracle . Normally, each call to the oracle returns a single classified example. They show that to model a system with multiple learners, each call to the oracle should return a summary of the examples seen by a single agent. Cesa Bianchi et al. 1995) also ....

M. Kearns and H. S. Seung (1995). Learning from a Population of Hypotheses, Machine Learning, 18:255-276.

....that the two robots move synchronously and share a polynomial length random string, then no communication is necessary. Thus with only minor modifications, our algorithms may be used in a distributed setting. Previous results showing the power of team learning are plentiful (e.g. CBF 93] KS93] DP 91] particularly in the field of inductive inference (see Smith [Smi94] for an excellent survey) There are also many results on learning unknown graphs under various conditions (e.g. BRS93] DP90] RS87] RS93] Rabin proposed the idea of dropping pebbles to mark nodes ....

Michael J. Kearns and H. Sebastian Seung. Learning from a population of hypotheses. In Proceedings of COLT '93, pages 101-- 110, 1993.

.... 93] consider the task of learning a probabilistic binary sequence given the predictions of a set of experts on the same sequence. They show how to combine the prediction strategies of several experts to predict nearly as well as the best of the experts. In a related paper, Kearns and Seung [KS93] explore the statistical problems of combining several independent hypotheses to learn a target concept from a known, restricted concept class. In their model, each hypothesis is learned from a different, independently drawn set of random examples, so the learner can combine the results to ....

Michael J. Kearns and H. Sebastian Seung. Learning from a population of hypotheses. In Proceedings of COLT '93, pages 101--110, 1993.

....curves in all datasets. Even for IB1 in the protein coding dataset (in Figure 11(a) when the complete learning curve is not observed, this phenomenon still occurs. This empirical result seems to be stronger, in terms of the expected performance of theory combination, than a theoretical result (Kearns Seung, 1995) based on the same working assumption. This theoretical work seeks . the possibility of somehow combining the independent hypotheses in a way that considerably outperforms any single hypothesis (Kearns Seung, 1995) The single hypothesis refers to any of the theory combination s ....

....of the expected performance of theory combination, than a theoretical result (Kearns Seung, 1995) based on the same working assumption. This theoretical work seeks . the possibility of somehow combining the independent hypotheses in a way that considerably outperforms any single hypothesis (Kearns Seung, 1995). The single hypothesis refers to any of the theory combination s constituent theories. Our result indicates that theory combination can significantly outperform not only its constituent theories but the theory learned from aggregating the available data; in spite of the the fact that this ....

Kearns, M. & H.S. Seung (1995), Learning from a Population of Hypotheses, Machine Learning, 18, pp. 255-276, Kluwer Academic Publishers.

No context found.

Kearns, M. and Seung, S. (1995). Learning from a Population of Hypotheses. Machine Learning, 18, 255-276

No context found.

Kearns, M., and Seung, H., Learning from a population of hypotheses, Machine Learning, 18, Issue 2/3, pp. 255--276, 1995.

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