M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, Vol. 4, No. 2, pp. 305--318, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....is that in boosting, the bootstrap samples are a ected by the performance of the current model. In addition, the nal output is a weighted average of the di erent models. Active data selection Some of the statistical active learning methods can also be applied to this type of data selection [60]. A special case of integrated mean square error, called integrated squared bias is used as the criterion to select a subset from available data to improve learning performance. However, it is assumed that the data is noiseless. Data weighting guided by evolution In [37] a method to weight the ....

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4(2):305-318, 1993.

....a local asymmetrically weighted similarity metric (LASM) approach for data compression [9] is shown to have superior performance compared to conventional k NN classification based methods. Similar concepts of data reduction and locally varying models based on neural networks are discussed in [10], 11] 12] The classification based condensation methods are, how ever, specific to (i.e. dependent on) the classification tasks and the models (e.g. k NN, perceptron) used. Data condensation of more generic nature is performed by classical vector quantization methods [13] using a set of ....

M. Plutowski and H. White, "Selecting Concise Training Sets from Clean Data," IEEE Trans. Neural Networks, vol. 4, no. 2, pp. 305-318, 1993.

....the candidate set. Thus, as training progresses, the size of the candidate set decreases while the size of the actual training set grows. Examples of incremental learning algorithms are optimal experiment design [1] information based objective functions [11] integrated squared bias minimization [12], error selection [13, 17] and sensitivity analysis incremental learning [3] Selective learning, where the network selects at each selection interval a new training subset from the original candidate set. Selected patterns are not removed from the candidate set. At each selection interval, ....

Plutowski, M., White, H.: Selecting Concise Training Sets from Clean Data. IEEE Transactions on Neural Networks. 4(2) (1993) 305-318

....a quadratic programming approach [49] Research on incremental learning for feedforward neural networks is more abundant than for selective learning. Most current incremental learning techniques have their roots in information theory, adapting Fedorov s optimal experiment design for NN learning [33, 38, 50, 51, 52]. The different information theoretic incremental learning algorithms are very similar, and differ only in whether they consider only bias, only variance, or both bias and variance terms in their selection criteria. These approaches are computationally very expensive, due to the required inversion ....

Plutowski, M., White, H.: Selecting Concise Training Sets from Clean Data, IEEE Transactions on Neural Networks, 4(2), 1993, 305-318.

....as measure of pattern informativeness. Plutowski and White use the change in integrated squared bias to select patterns [14] Zhang [7] and R obel [5] use the error between the output and target as indication of pattern informativeness. For classi cation problems Cohn et al. [ and Hwang et al. [15] use the distance of patterns from decision boundaries as selection criterion. This paper presents the Sensitivity Analysis Incremental Learning Algorithm (SAILA) for function approximation problems, which uses pattern sensitivity information as measure of informativeness. First order derivatives ....

M Plutowski, H White, Selecting Concise Training Sets from Clean Data, IEEE Transactions on Neural Networks, 4(2), pp 305-318, 1993.

....[4] is used for this purpose. Several incremental learning algorithms have been developed, differing mainly in the measure of pattern informativeness. Most current incremental learning techniques have their roots in information theory, adapting Fedorov s optimal experiment design for NN learning [1, 6, 9, 10, 12]. The different information theoretic incremental learning algorithms are very similar, and differ only in whether they consider only bias, only variance, or both bias and variance terms to quantify pattern informativeness. Zhang [13] and Robel [11] define informativeness as a function of the ....

M Plutowski, H White, Selecting Concise Training Sets from Clean Data, IEEE Transactions on Neural Networks, 4(2), March 1993, pp 305-318.

....contain enough information to fix the network function f n not only at the training patterns, but on the domain Xof the target function f t . To achieve this we want to adapt the training set during training and employ the net function to decide which pattern should be chosen. Plutowski and White [6] have done some work on active pattern selection, but did not employ cross validation to assess the generalization properties obtained by the training set. In contrast to their algorithm, the dynamic pattern selection proposed here, achieves concise training sets by continually validating the ....

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4(2):305--318, 1993.

....in this group of algorithms: those that start with a very small subset of the complete training set and sequentially add training samples, i.e. the training set is grown, and those that delete training samples from the complete training set, i.e. the training set is pruned. Plutowski et al. [12, 13] propose an algorithm that sequentially adds new training samples to the training set. The prerequisite of this algorithm are clean, i.e. noisefree, training data. A new training sample is then added to the training set with the aim to maximize the expected decrement in square error that would ....

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4:305-318, 1993.

....k = k 1 Select the data for model updating Update the approximate model end if end for END Algorithm 1: Algorithm for evolutionary optimization with approximate models. There are a number of methods for data selection in neural network training, which are usually called active learning [34, 35, 36, 37]. However, all of these methods rely on sucient data in order to employ methods from statistics. In the problem outlined in this paper, data are sparse and their collection is computationally expensive. At the same time, information about the topology of the search space and even more importantly ....

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4(2):305-318, 1993.

....using the same format as Figure 1. For these higher frequency categories, uncertainty sampling clearly dominates relevance sampling over the entire range of sample sizes considered. 4 Summary Uncertainty sampling and relevance sampling are two of a range of possible active exemplar selection [2] approaches to choosing training data. Both are far superior to random sampling when limitations of time or money mean that only a fraction of a data set can be labeled for training classification rules. Our corrected results show less difference between uncertainty sampling and random sampling ....

Mark Plutowski and Halbert White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4(2):305--318, March 1993.

....the learner itself is responsible for acquiring the training set. Here, we assume it can iteratively select a new input x #possibly from a constrained set#, observe the resulting output y, and incorporate the new example # x; y#into its training set. This contrasts with related work by Plutowski and White #1993#, which is concerned with #ltering an existing data set. In our case, x may be thought of as a query, experiment, or action, depending on the research #eld and problem domain. The question we will be concerned with is howtochoose which x to try next. There are many heuristics for choosing x, ....

Plutowski, M., & White, H. #1993#. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4, 305#318.

....training. c fl1999 AI Access Foundation and Morgan Kaufmann Publishers. All rights reserved. Argamon Dagan There are two main types of active learning. The first uses membership queries, in which the learner constructs examples and asks a teacher to label them (Angluin, 1988; MacKay, 1992b; Plutowski White, 1993). While this approach provides proven computational advantages (Angluin, 1987) it is not always applicable since it is not always possible to construct meaningful and informative unlabeled examples for training. This difficulty may be overcome when a large set of unlabeled training data is ....

Plutowski, M., & White, H. (1993). Selecting concise training sets from clean data. IEEE Trans. on Neural Networks, 4 (2).

....Several approaches have been developed to derive criteria for selection of examples in di#erent sort of networks, running from simple nets as perceptrons #Kinzel and Ruj#an, 1990# and linear classi#ers #Jung and Opper, 1996# to multilayer feed forward networks with continuous outputs. Among many others #see Plutowski and White, 1993, for more references# we can mention the work of Cohn et al. #Cohn et al., 1994; Cohn, 1996# that use partially trained networks to determine regions of uncertainty in the environment from which the examples are selected. Plutowski and White assume that a large amount of data has been collected and ....

MA. # Plutowski, M. and White. H. 1993. Selecting Concise Training Sets from Clean Data. IEEE Transactions on Neural Networks 4 #2#, 305.

....les diffrents algorithmes Dans le cas d une action indirecte (voir paragraphe III.1.1 et la figure III.2) le rseau dispose dj d un ensemble d apprentissage. Un premier type d algorithme va alors consister construire l aide de cet ensemble un sous ensemble contenant moins d exemples [Rbel 94] Plutowski White 93] Zhang 94] Krogh Vedelsby 95] Le sous ensemble est construit de manire incrmentale#: les exemples sont ajouts un un. Dans un deuxime type d algorithme, on balaie tous les exemples de l ensemble mais on pondre leur influence#: plus un exemple apporte de l information, plus on l apprend. ....

....calculer et correspond au premier critre qui vient intuitivement l esprit. Dans [Zhang 94] cependant, le fait de choisir les exemples dont l erreur est la plus leve est obtenu dans le cadre de la thorie de l information. Un critre beaucoup plus sophistiqu, et complexe calculer, a t propos dans [Plutowski White 93]#; il consiste choisir dans l ensemble d apprentissage l exemple qui maximise la dcroissance de l erreur calcule sur tout l ensemble d apprentissage. On montre que les exemples slectionns de cette manire ne correspondent pas aux exemples ayant une erreur maximale. Dans [Plutowski et al. 95] ce ....

[Article contains additional citation context not shown here]

Mark Plutowski & Halbert White. Selecting Concise Training Sets from Clean Data. IEEE Trans. on Neural Networks, 4(2), mars 1993.

....learning samples. Moreover, in the training set are normally samples which are redundant and do not provide usefull classification information. For this reason, different investigations have been developed in order to examine the effect of sample presentation and selection on learning, 19] [20], 21] 22] We have presented the BM as a new technique which can be applied to the field of pattern recognition. Now, the use of BM to construct robust an efficient NN classifiers is presented. They are used to incrementally select the samples of the training set in such way that several ....

Plutowski , M. and H. White, "Selecting Concise Training Sets from Clean Data," IEEE Transactions on Neural Networks, vol. 4, pp. 305--318, 1993.

No context found.

Plutowski, Mark E., and Halbert White. 1993. "Selecting concise training sets from clean data." IEEE Transactions on Neural Networks. 4, 3, pp.305-318.

.... 25 examples, Plus or Minus 2 Mark Plutowski Garrison Cottrell Halbert White Institute for Neural Computation Department of Computer Science and Engineering Department of Economics University of California, San Diego La Jolla, CA 92093 Abstract We apply active exemplar selection (Plutowski White, 1991; 1993) to predicting a chaotic time series. Given a fixed set of examples, the method chooses a concise subset for training. Fitting these exemplars results in the entire set being fit as well as desired. The algorithm incorporates a method for regulating network complexity, automatically adding ....

....units. The method requires an order of magnitude fewer floating point operations than training on the entire set of examples, is significantly cheaper than two contending exemplar selection techniques, and suggests a simpler active selection technique that performs comparably. 1 Introduction Plutowski White (1991; 1993), have developed a method of active selection of training exemplars for network learning. Active selection uses information about the state of the network when choosing new exemplars. The approach uses the statistical sampling criterion Integrated Squared Bias (ISB) to derive a greedy selection ....

[Article contains additional citation context not shown here]

Plutowski, Mark E., and Halbert White. 1993. "Selecting concise training sets from clean data." To appear, IEEE Transactions on Neural Networks. 3, 1.

No context found.

Plutowski, Mark E., and Halbert White. 1993. "Selecting concise training sets from clean data." IEEE Transactions on Neural Networks. 4, 3, pp.305-318.

No context found.

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, Vol. 4, No. 2, pp. 305--318, 1993.

No context found.

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, Vol. 4, No. 2, pp. 305-318, 1993.

No context found.

M. Plutowski and H. White, "Selecting concise training sets from clean data," IEEE Trans. Neural Networks, vol. 4, no. 2, pp. 305--318, 1993.

No context found.

Plutowski, M., & White, H. (1993). Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4, 305-318.

No context found.

M. Plutowski and H. White. Selecting concise training sets from clean data. IEEE Transactions on Neural Networks, 4(2):305--318, 1993.

No context found.

M. Plutowski and H. White, "Selecting concise training sets from clean data," IEEE Trans. Neural Networks, vol. 4, no. 2, pp. 305--318, 1993.

No context found.

M. Plutowski and H. White, "Selecting concise training sets from clean data," IEEE Trans. Neural Networks, vol. 4, no. 2, pp. 305--318, 1993.

First 50 documents

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