Combination of Classifiers for Supervised Learning: A Survey Sheng Ma IBM TJ Watson Research Cent... more Combination of Classifiers for Supervised Learning: A Survey Sheng Ma IBM TJ Watson Research Center Hawthorne, NY 10532 E-mail: shengma@ us. ibm. com Chuanyi Ji De. pt. of ECSE Rensselaer Polytechnic Institute Troy, NY 12180 E-mail: chuanyi@ ecse. rpi. edu Contents 1 Introduction 2 Background: Performance and Efficiency 2.1 Notation 2.2 Performance 2.3 Performance Issues 2.3. 1 Bias and Variance Dilemma.... 2.3. 2 Bias-variance and Local Minima. 3 Combinations of Well-Trained Models 3.1 Key Issues 3.2 Obtaining Base Models 3.3 Combination 3.4 ...
This work discovers that although network traffic has a complicated short- and long-range tempora... more This work discovers that although network traffic has a complicated short- and long-range temporal dependence, the corresponding wavelet coefficients are no longer long-range dependent. Therefore, a “short-range” dependent process can be used to model network traffic in the wavelet domain. Both independent and Markov models are investigated. Theoretical analysis shows that the independent wavelet model is sufficiently accurate in terms of the buffer overflow probability for fractional Gaussian noise traffic. Any model which captures additional correlations in the wavelet domain only improves the performance marginally. The independent wavelet model is then used as a unified approach to model network traffic including VBR MPEG video and Ethernet data. The computational complexity is O(N) for developing such wavelet models and generating synthesised traffic of length N, which is among the lowest attained.
IEEE INFOCOM '99. Conference on Computer Communications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat. No.99CH36320), 1999
A fast training algorithm is developed for two-layer feedforward neural networks based on a proba... more A fast training algorithm is developed for two-layer feedforward neural networks based on a probabilistic model for hidden representations and the EM algorithm. The algorithm decomposes training the original two-layer networks into training a set of single neurons. The individual neurons are then trained via a linear weighted regression algorithm. Significant improvement on training speed has been made using this algorithm for several bench-mark problems. Copyright 1997 Elsevier Science Ltd. All Rights Reserved.
ABSTRACT Clustering suffers from the curse of dimensionality, and similarity func-tions that use ... more ABSTRACT Clustering suffers from the curse of dimensionality, and similarity func-tions that use all input features with equal relevance may not be effective. We introduce an algorithm that discovers clusters in subspaces spanned by different combinations of dimensions via local weightings of features. This approach avoids the risk of loss of information encountered in global dimensionality reduction techniques, and does not assume any data dis-tribution model. Our method associates to each cluster a weight vector, whose values capture the relevance of features within the correspond-ing cluster. We experimentally demonstrate the gain in perfomance our method achieves, using both synthetic and real data sets. In particular, our results show the feasibility of the proposed technique to perform si-multaneous clustering of genes and conditions in microarray data.
In our previous work, we showed empirically that independent (Haar) wavelet models were parsimoni... more In our previous work, we showed empirically that independent (Haar) wavelet models were parsimonious, computationally efficient and accurate in modeling heterogeneous network traffic measured by both auto-covariance functions and buffer loss rate. We also proved analytically that such models were capable of capturing any decay rate of auto-covariance functions at large lags. But the simplicity of independent Haar wavelet models also results in deviations from the actual auto-covariance functions in small lags ...
Combination of Classifiers for Supervised Learning: A Survey Sheng Ma IBM TJ Watson Research Cent... more Combination of Classifiers for Supervised Learning: A Survey Sheng Ma IBM TJ Watson Research Center Hawthorne, NY 10532 E-mail: shengma@ us. ibm. com Chuanyi Ji De. pt. of ECSE Rensselaer Polytechnic Institute Troy, NY 12180 E-mail: chuanyi@ ecse. rpi. edu Contents 1 Introduction 2 Background: Performance and Efficiency 2.1 Notation 2.2 Performance 2.3 Performance Issues 2.3. 1 Bias and Variance Dilemma.... 2.3. 2 Bias-variance and Local Minima. 3 Combinations of Well-Trained Models 3.1 Key Issues 3.2 Obtaining Base Models 3.3 Combination 3.4 ...
This work discovers that although network traffic has a complicated short- and long-range tempora... more This work discovers that although network traffic has a complicated short- and long-range temporal dependence, the corresponding wavelet coefficients are no longer long-range dependent. Therefore, a “short-range” dependent process can be used to model network traffic in the wavelet domain. Both independent and Markov models are investigated. Theoretical analysis shows that the independent wavelet model is sufficiently accurate in terms of the buffer overflow probability for fractional Gaussian noise traffic. Any model which captures additional correlations in the wavelet domain only improves the performance marginally. The independent wavelet model is then used as a unified approach to model network traffic including VBR MPEG video and Ethernet data. The computational complexity is O(N) for developing such wavelet models and generating synthesised traffic of length N, which is among the lowest attained.
IEEE INFOCOM '99. Conference on Computer Communications. Proceedings. Eighteenth Annual Joint Conference of the IEEE Computer and Communications Societies. The Future is Now (Cat. No.99CH36320), 1999
A fast training algorithm is developed for two-layer feedforward neural networks based on a proba... more A fast training algorithm is developed for two-layer feedforward neural networks based on a probabilistic model for hidden representations and the EM algorithm. The algorithm decomposes training the original two-layer networks into training a set of single neurons. The individual neurons are then trained via a linear weighted regression algorithm. Significant improvement on training speed has been made using this algorithm for several bench-mark problems. Copyright 1997 Elsevier Science Ltd. All Rights Reserved.
ABSTRACT Clustering suffers from the curse of dimensionality, and similarity func-tions that use ... more ABSTRACT Clustering suffers from the curse of dimensionality, and similarity func-tions that use all input features with equal relevance may not be effective. We introduce an algorithm that discovers clusters in subspaces spanned by different combinations of dimensions via local weightings of features. This approach avoids the risk of loss of information encountered in global dimensionality reduction techniques, and does not assume any data dis-tribution model. Our method associates to each cluster a weight vector, whose values capture the relevance of features within the correspond-ing cluster. We experimentally demonstrate the gain in perfomance our method achieves, using both synthetic and real data sets. In particular, our results show the feasibility of the proposed technique to perform si-multaneous clustering of genes and conditions in microarray data.
In our previous work, we showed empirically that independent (Haar) wavelet models were parsimoni... more In our previous work, we showed empirically that independent (Haar) wavelet models were parsimonious, computationally efficient and accurate in modeling heterogeneous network traffic measured by both auto-covariance functions and buffer loss rate. We also proved analytically that such models were capable of capturing any decay rate of auto-covariance functions at large lags. But the simplicity of independent Haar wavelet models also results in deviations from the actual auto-covariance functions in small lags ...
Uploads