Single-layer learning revisited: a stepwise procedure for building and training a neural network

Knerr, Stefan; Personnaz, L.; Dreyfus, Gérard

doi:10.1007/978-3-642-76153-9_5

Cited by 576 publications

(302 citation statements)

References 6 publications

Supporting

Mentioning

294

Contrasting

Unclassified

Order By: Relevance

“…Pairwise methods have previously been shown to improve performance for neural networks (Knerr et al 1990(Knerr et al , 1992Price et al 1995;Lu and Ito 1999), support vector machines (Schmidt and Gish 1996;Hastie and Tibshirani 1998;Kreßel 1999;Hsu and Lin 2002), statistical learning (Bradley and Terry 1952;Friedman 1996), rule and decision tree learning (Fürnkranz 2002(Fürnkranz , 2003 and others. Moreover, for two of the datasets used in this study, we have also experimented with calibrated pairwise ranking using support vector machines as base classifiers, and obtained similar results (Brinker et al 2006).…”

Section: Learning Algorithmsmentioning

confidence: 99%

Multilabel classification via calibrated label ranking

et al. 2008

View full text Add to dashboard Cite

Label ranking studies the problem of learning a mapping from instances to rankings over a predefined set of labels. Hitherto existing approaches to label ranking implicitly operate on an underlying (utility) scale which is not calibrated in the sense that it lacks a natural zero point. We propose a suitable extension of label ranking that incorporates the calibrated scenario and substantially extends the expressive power of these approaches. In particular, our extension suggests a conceptually novel technique for extending the common learning by pairwise comparison approach to the multilabel scenario, a setting previously not being amenable to the pairwise decomposition technique. The key idea of the approach is to introduce an artificial calibration label that, in each example, separates the relevant from the irrelevant labels. We show that this technique can be viewed as a combination of pairwise preference learning and the conventional relevance classification technique, where a separate classifier is trained to predict whether a label is relevant or not. Empirical results in the area of text categorization, image classification and gene analysis underscore the merits of the calibrated model in comparison to state-of-the-art multilabel learning methods.

show abstract

Section: Learning Algorithmsmentioning

confidence: 99%

Multilabel classification via calibrated label ranking

et al. 2008

View full text Add to dashboard Cite

show abstract

“…The 'One-against-one' [7] multiclass approach is implemented in this study as it has been shown to have comparable if not better generalized accuracy than alternative techniques and requires considerably less training time [8], [9]. The method consists of constructing an SVM for each pair of classes.…”

Section: Support Vector Machinesmentioning

confidence: 99%

Steady State RF Fingerprinting for Identity Verification: One Class Classifier versus Customized Ensemble

Kroon¹,

Bergin²,

Kennedy³

et al. 2010

Artificial Intelligence and Cognitive Science

View full text Add to dashboard Cite

Abstract.Mobile phone proliferation and increasing broadband penetration presents the possibility of placing small cellular base stations within homes to act as local access points. This can potentially lead to a very large increase in authentication requests hitting the centralized authentication infrastructure unless access is mediated at a lower protocol level. A study was carried out to examine the effectiveness of using Support Vector Machines to accurately identify if a mobile phone should be allowed access to a local cellular base station using differences imbued upon the signal as it passes through the analogue stages of its radio transmitter. Whilst allowing prohibited transmitters to gain access at the local level is undesirable and costly, denying service to a permitted transmitter is simply unacceptable. Two different learning approaches were employed, the first using One Class Classifiers (OCCs) and the second using customized ensemble classifiers. OCCs were found to perform poorly, with a true positive (TP) rate of only 50% (where TP refers to correctly identifying a permitted transmitter) and a true negative (TN) rate of 98% (where TN refers to correctly identifying a prohibited transmitter). The customized ensemble classifier approach was found to considerably outperform the OCCs with a 97% TP rate and an 80% TN rate.

show abstract

“…Classification of a test pattern is done according to the maximum output of these ten classifiers. There are some other ways to combine many two-class classifiers into a multiclass classifier (Platt, Cristianini, & Shawe Taylor, in Press;Friedman, 1996;Knerr, Personnaz, & Dreyfus, 1990).…”

Section: Experimental Settingsmentioning

confidence: 99%

Untitled

Long

2002

Machine Learning

132

View full text Add to dashboard Cite

Abstract. We describe a new incremental algorithm for training linear threshold functions: the Relaxed Online Maximum Margin Algorithm, or ROMMA. ROMMA can be viewed as an approximation to the algorithm that repeatedly chooses the hyperplane that classifies previously seen examples correctly with the maximum margin. It is known that such a maximum-margin hypothesis can be computed by minimizing the length of the weight vector subject to a number of linear constraints. ROMMA works by maintaining a relatively simple relaxation of these constraints that can be efficiently updated. We prove a mistake bound for ROMMA that is the same as that proved for the perceptron algorithm. Our analysis implies that the maximum-margin algorithm also satisfies this mistake bound; this is the first worst-case performance guarantee for this algorithm. We describe some experiments using ROMMA and a variant that updates its hypothesis more aggressively as batch algorithms to recognize handwritten digits. The computational complexity and simplicity of these algorithms is similar to that of perceptron algorithm, but their generalization is much better. We show that a batch algorithm based on aggressive ROMMA converges to the fixed threshold SVM hypothesis.

show abstract

Single-layer learning revisited: a stepwise procedure for building and training a neural network

Cited by 576 publications

References 6 publications

Multilabel classification via calibrated label ranking

Multilabel classification via calibrated label ranking

Steady State RF Fingerprinting for Identity Verification: One Class Classifier versus Customized Ensemble

Untitled

Contact Info

Product

Resources

About