J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--276, Orlando, Fl., April 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....more than one network can be fused to provide a more reliable classification decision. A multitude of combining methods have emerged recently, along with a 20 large number of applications. Ghosh et al. look at different types of combiners and apply the methods to underwater sonar classification [59, 62]. Hansen and Salamon show that ensemble methods improve classification through a majority vote for independent classifiers [70] Hampshire and Waibel use the normalized sum of three networks trained on different objective functions for speech recognition, and propose a modular structure where ....

J. Ghosh, S. Beck, and C. Chu, Evidence combination techniques for robust classification of short-duration oceanic signals, in SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, Orlando, Fl., April 1992, pp. 266--276.

....[ 1989 ] In the field of neural networks, a number of researchers have looked at the advantages of combining multiple predictions. Lincoln and Skrzypek [ 1990 ] and Hansen and Salamon [ 1990 ] both explore the advantages of combining groups of networks in a simple way. Many researchers [ Ghosh et al. 1992; Hashem et al. 1993; Perrone and Cooper, 1994; Rogova, 1994; Wolpert, 1992 ] have studied the problem of combining predictions in a more robust way, taking 5 The main difficulty in pursuing these other experiments is CPU time; our protein folding experiments required us to train 500 networks ....

J. Ghosh, L. Deuser, and S. Beck. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, volume 1706, pages 266--276, Orlando, FL, 1992.

....a 1 of M encoding of the output units summing to one, estimate Bayesian a posteriori probabilities. This allows to see the outputs of an neural network not only as crisp classification results like in the majority vote algorithm, but to use additional probability information of the outputs. In [6] a combination algorithm, called the Entropy approach, based on this information is presented. For every sample, the function H(c) is calculated and the class with max(H(c) is selected. H(c) 1 n n X i=1 o c;i Gamma P c o c;i ln o c;i (3) 4.3 Related Works Opitz and Shavlik [8] ....

J. Ghosh, S. Beck, and C.-C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Science of Artificial Neural Networks (SPIE Vol. 1710), pages 266--276, 1992. http://www.lans.ece.utexas.edu.

.... abnormal, and artifact. Class confidence values are derived using a method similar to that used by Shlien [86] which are combined by a heuristic method and a Dempster Shafer theory based method. Combination results show considerable improvements in classification correctness. Ghosh et al. [27] present and compare four methods of evidence combination for classification of underwater acoustic signals. The four methods are entropy based weighting of individual classifier outputs, combination of confidence factors in a manner similar to that used in MYCIN [10] majority voting, and ....

J. Ghosh, S. D. Beck, and C.-C. Chu, "Evidence combination techniques for robust classification of short-duration oceanic signals," SPIE Vol. 1706---Adaptive and Learning Systems, 266--276, April 1992.

.... outputs of individual classifiers has been suggested by different researchers as an alternative to selecting the best network [8, 9, 10] Methods that select the class with the highest activation value, use the geometric mean or entropy based criteria, or perform a majority vote have been analyzed [11, 12, 13]. Methods based on confidence factors obtained through the theory evidence have also been studied [14] Weighted averaging has been proposed, along with different methods of computing the proper classifier weights [8, 9] A survey of leading combining techniques, along with experimental results is ....

.... based on confidence factors obtained through the theory evidence have also been studied [14] Weighted averaging has been proposed, along with different methods of computing the proper classifier weights [8, 9] A survey of leading combining techniques, along with experimental results is given in [11, 12]. Combining techniques such as majority voting can generally be applied to any type of classifier, while others rely on specific outputs, or specific interpretations of the output. For example, the confidence factors method relies on the interpretation of the outputs as the belief that the ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--276, Orlando, Fl., April 1992.

.... reduce the bias and make the classifier more robust [7] In the context of supervised feedforward networks, interpretation of network outputs as Bayesian a posteriori probabilities [8] provides a sound basis for combining the results from multiple classifiers to yield more accurate classification [1, 3, 4]. The concept of stacked generalization, an inductive approach to combining generalizers, has been recently introduced by Wolpert [9] A framework for hybrid neural networks in regression estimates was discussed in [7] In this paper, we focus on the statistical aspects of such combining methods. ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--76, Orlando, Fl., April 1992.

....Data In order to examine the benefits of combining and the effect of correlation on combining results, we use a difficult data set extracted from underwater acoustic signals. From the original passive sonar returns from four different underwater objects, a 25 dimensional feature set was extracted (Ghosh et al. 1992; Ghosh et al. 1996) Each patterns consists of 16 Gabor wavelet coefficients, 8 temporal descriptors and spectral measurements and 1 value denoting signal duration. There were 496 patterns in the training set, and 823 in the test set (Table I) The data is available at URL ....

Ghosh, J., Beck, S., and Chu, C. (1992). Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--276.

...., the output of the combiner. beliefs in the Dempster Shafer sense are also available [37, 39, 50, 51] Combiners have also been successfully applied a multitude of real world problems [5, 7, 17, 25, 41, 52] A survey of leading combining techniques, along with experimental results is given in [15, 17]. Combining techniques such as majority voting can generally be applied to any type of classifier, while others rely on specific outputs, or specific interpretations of the output. For example, the confidence factors method found in machine learning literature relies on the interpretation of the ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--276, Orlando, Fl., April 1992. 31

....such as those involving very noisy data, limited number of training data, or unusually high dimensional patterns. Several combining methods have proved effective in improving the classifier performance, including simple averaging, majority voting, weighted averaging or evidence based combining [8, 9, 10, 11, 12, 13, 14]. A survey of leading combining techniques, along with experimental results is given in [8, 9] Linear combining techniques for regression (or function approximation) problems have been mathematically analyzed [15, 16] However, despite the increasing body of experimental results showing ....

.... Several combining methods have proved effective in improving the classifier performance, including simple averaging, majority voting, weighted averaging or evidence based combining [8, 9, 10, 11, 12, 13, 14] A survey of leading combining techniques, along with experimental results is given in [8, 9]. Linear combining techniques for regression (or function approximation) problems have been mathematically analyzed [15, 16] However, despite the increasing body of experimental results showing classification improvements due to combining, there has been no analytical study that can quantify the ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--276, Orlando, Fl., April 1992.

....patterns, at least in regions where there are sufficient training patterns [7, 8] Interpretation of network outputs as Bayesian probabilities provides a sound basis for combining the results from multiple classifiers to yield more accurate classification. This idea was first explored by us in [3] wherein we examined several techniques for combining such evidences , and compared their efficacies for classifying underwater acoustic signals. In this paper, we focus again on passive sonar signals. This problem domain is of special interest because of the high dimensionality of the input ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of shortduration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--76, Orlando, Fl., April 1992.

....Section 3 describes the data set representing 6 classes of radar signals. Experimental results obtained by the classifiers on this data set are presented in Section 4. Classification results for short duration oceanic signals represented by 25 dimensional vectors, are given in a companion paper [8]. 2 Overview of ANN Classifiers Used 2.1 MLP with Weight Decay The first candidate used is the standard fully connected MLP network that adapts weights using gradient descent of the mean squared error (MSE) in weight space. Since such backprop networks are widely known and are perhaps the ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--76, Orlando, Fl., April 1992.

....much of the training data does not participate in determining the network parameter values. We note that this drawback is even more severe when the number of training samples is limited. For all these reasons, the MLP is not considered further in this paper, and instead the reader is referred to [52] for our results on using an MLP with optimal brain damage [9] for classifying underwater signals from biologic sources. 3.3 Efficient Adaptive Kernel Classifiers We first summarize the LVQ and RBF procedures and then introduce two hybrid networks that attempt to incorporate the best of both LVQ ....

....of the 19 deviant signals, the summation of Eq. 18 is less than 0.5, strongly indicating that these signals do not resemble any signal in the training set. Based on the interpretation of the outputs as aposterior class probabilities, two methods for evidence combination were proposed and used [52]: 1. Entropy Based Integrator. In this method, a weighted average of the outputs of n different classifiers is first performed, with a larger entropy resulting in a smaller weight. The integrator then selects the class corresponding to the maximum value, providing this value is above a threshold. ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--76, Orlando, Fl., April 1992.

....and multisensor fusion in pattern recognition. In the context of feedforward networks, interpretation of network outputs as Bayesian probabilities provides a sound basis for combining the results from multiple classifiers to yield more accurate classification. This idea was first explored by us in [GBC92] wherein we examined several techniques for combining such evidences , and compared their efficacies for classifying underwater acoustic signals. The results clearly showed that using multiple classifiers provides better and more robust classification decisions. Multiple classifier integration ....

....classifiers for more accurate and robust results. The work presented in this paper is part of a larger project on the design of a detection and classification system that uses a hybrid of ANN and statistical pattern recognition techniques tailored to recognizing short duration oceanic signals [Gho91, GBC92, GDB92, BG92]. 2 Overview of ANN Classifiers Used Our experiences, corroborated by those of several other researchers (see [NL91] for example) show that classification error rates are similar across different ANN classifiers when they are powerful enough to form minimum error decision regions, if they are ....

J. Ghosh, S. Beck, and C.C. Chu. Evidence combination techniques for robust classification of short-duration oceanic signals. In SPIE Conf. on Adaptive and Learning Systems, SPIE Proc. Vol. 1706, pages 266--76, Orlando, Fl., April 1992.

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