## Integration Of Neural Classifiers For Passive Sonar Signals (1995)

Venue: | In C.T. Leondes, editor, DSP Theory and Applications |

Citations: | 15 - 12 self |

### BibTeX

@INPROCEEDINGS{Ghosh95integrationof,

author = {Joydeep Ghosh and Kagan Tumer and Steven Beck and Larry Deuser},

title = {Integration Of Neural Classifiers For Passive Sonar Signals},

booktitle = {In C.T. Leondes, editor, DSP Theory and Applications},

year = {1995},

pages = {301--338},

publisher = {Academic Press}

}

### OpenURL

### Abstract

The identification and classification of underwater acoustic signals is an extremely difficult problem because of low SNRs and a high degree of variability in the signals emanated from the same type of sound source. Since different classification techniques have different inductive biases, a single method cannot give the best results for all signal types. Rather, more accurate and robust classification can obtained by combining the outputs (evidences) of multiple classifiers based on neural network and/or statistical pattern recognition techniques. In this paper, five approaches are compared for integrating the decisions made by networks using sigmoidal activation functions exhibiting global responses with those made by localized basis function networks. These methods are compared using realistic oceanic data. The first method uses an entropy-based weighting of individual classifier outputs. The second is based on combination of confidence factors in a manner similar to that used in MY...

### Citations

3921 |
Pattern Classification and Scene Analysis
- Duda, Hart
- 1973
(Show Context)
Citation Context ...t classes have substantial overlap in their density distributions, then the statistically optimal expected correct classification achieved by Bayes decision theory can be significantly less than 100% =-=[DH73]-=-. For real-life, difficult signals such as the underwater acoustic signals considered in this paper, the apriori class distributions are not known, and thus Bayesian a posteriori probabilities are not... |

555 | Stacked Generalization
- Wolpert
- 1992
(Show Context)
Citation Context ...using an approach based on the theory of binary decision trees [Shl90]. The concept of stacked generalization, an inductive approach to combining generalizers, has been recently introduced by Wolpert =-=[Wol92]-=-. Another aspect of getting more robust classification results when using a gradient-descent based network such as the MLP is explored in [KL91]. Their work uses a bank of MLPs that are identical exce... |

476 | Fast learning in networks of locally-tuned processing units - Moody, Darken - 1989 |

434 | Multivariable functional interpolation and adaptive networks - Broomhead, Lowe |

419 | Optimal brain damage
- LeCun, Denker, et al.
- 1990
(Show Context)
Citation Context ...n MLP which is trained using the back-propagation rule with momentum. A decaying learning rate is used. Though we have also experimented with various pruning techniques such as "Optimal Brain Dam=-=age" [CDS90]-=-, the results were not statistically different from a fully connected MLP of appropriate size, and hence are not reported here. 2.2 Localized Basis Function Networks. The second network used for this ... |

365 | Computer Systems that Learn - Weiss, Kulikowski - 1995 |

267 | Neural network classifiers estimate Bayesian a posteriori probabiüties - D, Lippmann - 1991 |

97 |
A model of inexact reasoning in medicine
- Shortliffe, Buchanan
- 1975
(Show Context)
Citation Context ...llel combination of rules in expert systems. Certainty factors were introduced in the MYCIN expert system for reasoning in expert systems under uncertainty, and reflect the confidence in a given rule =-=[SB75]-=-. The original method of rule combination in MYCIN was later expressed in a more probabilistic framework by Heckerman [Hec86], and serves as the basis for the method proposed below: First, the outputs... |

74 |
Hybrid System for Protein Secondary Structure Prediction
- Zhang
- 1992
(Show Context)
Citation Context ...pose estimation. Even the direct use of a second MLP network to combine the outputs of diverse classifiers has shown statistically significant improvement in prediction of protein secondary structure =-=[ZMW92]-=-. 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 patterns, and the non-trivial decision boundaries d... |

60 |
Pattern classification using neural networks
- Lippmann
- 1989
(Show Context)
Citation Context ...Artificial neural network (ANN) approaches to problems in the field of pattern recognition and signal processing have led to the development of various "neural" classifiers using feedforward=-= networks [Lip89]-=-. These include the Multi-Layer Perceptron (MLP) as well as kernelbased classifiers such as those employing Radial or Elliptical Basis Functions (EBFs) [BL88, MD89]. These networks can serve as non-pa... |

50 | A probabilistic approach to the understanding and training of neural network classi ers - Gish - 1990 |

39 |
An inference technique for integrating knowledge from disparate sources. These proceedings
- Garvey, Lowrance, et al.
(Show Context)
Citation Context ...egrating knowledge from disparate sources has been explored in several fields. Particularly notable are the Dempster-Shaefer theory of evidence and evidence combination in machine learning literature =-=[GLF81]-=-, 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 fr... |

28 |
Practical characteristics of neural network and conventional pattern classi ers on arti cial and speech problems
- Lee, Lippmann
- 1990
(Show Context)
Citation Context ...hniques 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 properly tuned, and w... |

25 | A neural network based hybrid system for detection, characterization, and classification of short-duration oceanic signals - Ghosh, Deuser, et al. - 1992 |

24 |
Predicting the future: Advantages of semilocal units
- Hartman, Keeler
- 1991
(Show Context)
Citation Context ...ons, they are not so practical and efficient for high dimensional inputs. This is because a much larger number of kernel nodes are required to cover higher dimensional spaces with adequate resolution =-=[HK91]-=-. Moreover, they are more sensitive to noise and to irrelevant inputs [BG92]. The abovementioned drawbacks can be countered by using Elliptical Basis Function (EBF) networks that employ Gaussian basis... |

22 | Mapping neural networks onto messagepassing multicomputers
- Ghosh, Hwang
- 1989
(Show Context)
Citation Context ...roperly tuned, and when sufficient training data is available. Practical characteristics such as training time, classification time and memory requirements, however, can differ by orders of magnitude =-=[GH89]-=-. Also, the classifiers differ in their robustness against noise, effects of small training sets, and in their ability to handle high-dimensional inputs [BG92]. A good review of probabilistic, hyperpl... |

18 | Evidence combination techniques for robust classification of short-duration oceanic signals
- Ghosh, Beck, et al.
- 1992
(Show Context)
Citation Context ...on 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 w-=-e examined several techniques for combining such "evidences", and compared their efficacies for classifying underwater acoustic signals. The results clearly showed that using multiple classi... |

14 |
Multiple binary decision tree classifiers
- Shlien
- 1990
(Show Context)
Citation Context ...more robust classification decisions. Multiple classifier integration has also been suggested for automated cytology screening [LHDN91], using an approach based on the theory of binary decision trees =-=[Shl90]-=-. The concept of stacked generalization, an inductive approach to combining generalizers, has been recently introduced by Wolpert [Wol92]. Another aspect of getting more robust classification results ... |

13 | Least squares learning and approximation of posterior probabilities on classification problems by neural network modcls - Shoemaker, Carlin, et al. - 1991 |

11 |
J.(1992)-Pattern recognition properties of neural networks
- Makhoul
(Show Context)
Citation Context ...w values of MSE, f c (x) approximates P (c=x) according to Eq. 6. Let f c;i (x) be the output of the node that denotes membership in class c in the ith neural classifier. We expect that, for all i, x =-=[Mak91]-=-: X c f c;i ' 1: (7) Similarly, if the posteriori estimate is very good, one would expect for all c; i: 1 N N X j=1 f c;i (x j ) ' P (c); (8) where j indexes the N training data samples, and P(c) is o... |

10 |
Integration of neural networks and decision tree classifiers for automated cytology screening
- Lee, Hwang, et al.
- 1991
(Show Context)
Citation Context ...sults clearly showed that using multiple classifiers provides better and more robust classification decisions. Multiple classifier integration has also been suggested for automated cytology screening =-=[LHDN91]-=-, using an approach based on the theory of binary decision trees [Shl90]. The concept of stacked generalization, an inductive approach to combining generalizers, has been recently introduced by Wolper... |

9 |
A hybrid neural network classifier of short duration acoustic signals
- Beck, Deuser, et al.
- 1991
(Show Context)
Citation Context ... 284 2 Ice 175 175 3 Whale Sound 1 129 129 4 Whale Sound 2 118 235 Total 564 823 Signal preprocessing as well as the extraction of good feature vectors is crucial to the performance of the classifiers=-=[BDSW91]-=-. For FS1, the 25-dimensional feature vectors extracted from the raw signals consist of : ffl 16 coefficients of Gabor wavelets---a multiscale representation that does not assume signal stationarity[C... |

8 |
Noise sensitivity of static neural classifiers
- Beck, Ghosh
- 1992
(Show Context)
Citation Context ...wever, can differ by orders of magnitude [GH89]. Also, the classifiers differ in their robustness against noise, effects of small training sets, and in their ability to handle high-dimensional inputs =-=[BG92]-=-. A good review of probabilistic, hyperplane, kernel and exemplar-based classifiers that discusses the relative merit of various schemes within each category, is available in [Lip89, NL91]. Comparison... |

8 |
Probabilistic interpretation for MYCIN’s uncertainty factors
- Heckerman
- 1986
(Show Context)
Citation Context ...xpert systems under uncertainty, and reflect the confidence in a given rule [SB75]. The original method of rule combination in MYCIN was later expressed in a more probabilistic framework by Heckerman =-=[Hec86]-=-, and serves as the basis for the method proposed below: First, the outputs, which are in the range [0,1], are mapped into confidence factors (CFs) in the range [-1,1] using a log transformation. Then... |

7 |
Multilayer Feedforward Potential Function Network
- Lee, Kil
- 1988
(Show Context)
Citation Context ...fferent dimensions. Thus, for an EBF network, Eq. 1 is used with R j (x p ) = e \Gamma 1 2 \Sigma k (x pk \Gammax jk ) 2 oe 2 jk : (2) The EBF network is a type of Gaussian Potential Function Network =-=[LK88]-=- which involves segmentation of the input domain into several potential fields in form of Gaussians. The Gaussian potential functions of this scheme need not be radially symmetric functions. Instead, ... |

6 |
Recognition and pose estimation of 3-D objects from a single 2-D perspective view by banks of neural networks
- Khotanzad, Liou
- 1991
(Show Context)
Citation Context ... generalizers, has been recently introduced by Wolpert [Wol92]. Another aspect of getting more robust classification results when using a gradient-descent based network such as the MLP is explored in =-=[KL91]-=-. Their work uses a bank of MLPs that are identical except that they use different initial values of weights, for 3-D object classification and pose estimation. Even the direct use of a second MLP net... |

5 | Efficient training procedures for adaptive kernel classifiers - Chakravarthy, Ghosh, et al. - 1991 |

3 | et al. Adaptive kernel classifiers for short-duration oceanic signals - Ghosh - 1991 |