Arrow Research search

Author name cluster

Matthew Norton

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

2 papers
1 author row

Possible papers

2

AAAI Conference 2019 Conference Paper

Generalized Batch Normalization: Towards Accelerating Deep Neural Networks

  • Xiaoyong Yuan
  • Zheng Feng
  • Matthew Norton
  • Xiaolin Li

Utilizing recently introduced concepts from statistics and quantitative risk management, we present a general variant of Batch Normalization (BN) that offers accelerated convergence of Neural Network training compared to conventional BN. In general, we show that mean and standard deviation are not always the most appropriate choice for the centering and scaling procedure within the BN transformation, particularly if ReLU follows the normalization step. We present a Generalized Batch Normalization (GBN) transformation, which can utilize a variety of alternative deviation measures for scaling and statistics for centering, choices which naturally arise from the theory of generalized deviation measures and risk theory in general. When used in conjunction with the ReLU non-linearity, the underlying risk theory suggests natural, arguably optimal choices for the deviation measure and statistic. Utilizing the suggested deviation measure and statistic, we show experimentally that training is accelerated more so than with conventional BN, often with improved error rate as well. Overall, we propose a more flexible BN transformation supported by a complimentary theoretical framework that can potentially guide design choices.

PDF Details

JMLR Journal 2017 Journal Article

Soft Margin Support Vector Classification as Buffered Probability Minimization

  • Matthew Norton
  • Alexander Mafusalov
  • Stan Uryasev

In this paper, we show that the popular C-SVM, soft-margin support vector classifier is equivalent to minimization of Buffered Probability of Exceedance (bPOE), a recently introduced characterization of uncertainty. To show this, we introduce a new SVM formulation, called the EC-SVM, which is derived from a simple bPOE minimization problem that is easy to interpret with a meaningful free parameter, optimal objective value, and probabilistic derivation. Over the range of its free parameter, the EC-SVM has both a convex and non-convex case which we connect to existing SVM formulations. We first show that the C-SVM, formulated with any regularization norm, is equivalent to the convex EC-SVM. Similarly, we show that the E$\nu$-SVM is equivalent to the EC-SVM over its entire parameter range, which includes both the convex and non-convex case. These equivalences, coupled with the interpretability of the EC-SVM, allow us to gain surprising new insights into the C-SVM and fully connect soft margin support vector classification with superquantile and bPOE concepts. We also show that the EC-SVM can easily be cast as a robust optimization problem, where bPOE is minimized with data lying in a fixed uncertainty set. This reformulation allows us to clearly differentiate between the convex and non-convex case, with convexity associated with pessimistic views of uncertainty and non-convexity associated with optimistic views of uncertainty. Finally, we address some practical considerations. First, we show that these new insights can assist in making parameter selection more efficient. Second, we discuss optimization approaches for solving the EC-SVM. Third, we address the issue of generalization, providing generalization bounds for both bPOE and misclassification rate. [abs] [ pdf ][ bib ] &copy JMLR 2017. ( edit, beta )

PDF Details
v2026.09.13