Browsing by Author "Scott, Clayton D."

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Controlling False Alarms with Support Vector Machines
    (2006-05-01) Davenport, Mark A.; Baraniuk, Richard G.; Scott, Clayton D.; Digital Signal Processing (http://dsp.rice.edu/)
    We study the problem of designing support vector classifiers with respect to a Neyman-Pearson criterion. Specifically, given a user-specified level alpha, 0 < alpha < 1, how can we ensure a false alarm rate no greater than a while minimizing the miss rate? We examine two approaches, one based on shifting the offset of a conventionally trained SVM and the other based on the introduction of class-specific weights. Our contributions include a novel heuristic for improved error estimation and a strategy for efficiently searching the parameter space of the second method. We also provide a characterization of the feasible parameter set of the 2nu-SVM on which the second approach is based. The proposed methods are compared on four benchmark datasets.
  • Thumbnail Image
    Item
    Learning minimum volume sets with support vector machines
    (2006-09-01) Davenport, Mark A.; Baraniuk, Richard G.; Scott, Clayton D.
    Given a probability law P on d-dimensional Euclidean space, the minimum volume set (MV-set) with mass beta , 0 < beta < 1, is the set with smallest volume enclosing a probability mass of at least beta. We examine the use of support vector machines (SVMs) for estimating an MV-set from a collection of data points drawn from P, a problem with applications in clustering and anomaly detection. We investigate both one-class and two-class methods. The two-class approach reduces the problem to Neyman-Pearson (NP) classification, where we artificially generate a second class of data points according to a uniform distribution. The simple approach to generating the uniform data suffers from the curse of dimensionality. In this paper we (1) describe the reduction of MV-set estimation to NP classification, (2) devise improved methods for generating artificial uniform data for the two-class approach, (3) advocate a new performance measure for systematic comparison of MV-set algorithms, and (4) establish a set of benchmark experiments to serve as a point of reference for future MV-set algorithms. We find that, in general, the two-class method performs more reliably.
  • Thumbnail Image
    Item
    Minimax support vector machines
    (2007-08-01) Davenport, Mark A.; Baraniuk, Richard G.; Scott, Clayton D.
    We study the problem of designing support vector machine (SVM) classifiers that minimize the maximum of the false alarm and miss rates. This is a natural classification setting in the absence of prior information regarding the relative costs of the two types of errors or true frequency of the two classes in nature. Examining two approaches – one based on shifting the offset of a conventionally trained SVM, the other based on the introduction of class-specific weights – we find that when proper care is taken in selecting the weights, the latter approach significantly outperforms the strategy of shifting the offset. We also find that the magnitude of this improvement depends chiefly on the accuracy of the error estimation step of the training procedure. Furthermore, comparison with the minimax probability machine (MPM) illustrates that our SVM approach can outperform the MPM even when the MPM parameters are set by an oracle.
  • Thumbnail Image
    Item
    Regression level set estimation via cost-sensitive classification
    (2007-06-01) Scott, Clayton D.; Davenport, Mark A.
    Regression level set estimation is an important yet understudied learning task. It lies somewhere between regression function estimation and traditional binary classification, and in many cases is a more appropriate setting for questions posed in these more common frameworks. This note explains how estimating the level set of a regression function from training examples can be reduced to cost-sensitive classification. We discuss the theoretical and algorithmic benefits of this learning reduction, demonstrate several desirable properties of the associated risk, and report experimental results for histograms, support vector machines, and nearest neighbor rules on synthetic and real data.
  • Thumbnail Image
    Item
    Tuning support vector machines for minimax and Neyman-Pearson classification
    (2008-08-19) Scott, Clayton D.; Baraniuk, Richard G.; Davenport, Mark A.
    This paper studies the training of support vector machine (SVM) classifiers with respect to the minimax and Neyman-Pearson criteria. In principle, these criteria can be optimized in a straightforward way using a cost-sensitive SVM. In practice, however, because these criteria require especially accurate error estimation, standard techniques for tuning SVM parameters, such as crossvalidation, can lead to poor classifier performance. To address this issue, we first prove that the usual cost-sensitive SVM, here called the 2C-SVM, is equivalent to another formulation called the 2nu-SVM. We then exploit a characterization of the 2nu-SVM parameter space to develop a simple yet powerful approach to error estimation based on smoothing. In an extensive experimental study we demonstrate that smoothing significantly improves the accuracy of cross-validation error estimates, leading to dramatic performance gains. Furthermore, we propose coordinate descent strategies that offer significant gains in computational efficiency, with little to no loss in performance.