Dissimilarity representations allow for building good classifiers

Pękalska, Elżbieta; Duin, Robert P. W.

doi:10.1016/s0167-8655(02)00024-7

Cited by 193 publications

(130 citation statements)

References 5 publications

Supporting

Mentioning

128

Contrasting

Unclassified

Order By: Relevance

“…. , I n } by their pair-wise dissimilarities d(I i , I j ) and build classifiers on the obtained dissimilarity representation [8]. From the matrix of pair-wise image dissimilarities D = [d(I i , I j )] n×n computed from the set of images, there exist different ways of arriving at a feature vector space where traditional vector space methods can be applied.…”

Section: Image Dissimilarity Spacementioning

confidence: 99%

“…Further, problems where only a global image label is available are handled automatically since the classification is done at the image level. The images are mapped into a dissimilarity space [8] in which a standard vector space-based classifier can be directly applied, and the soft output of this classifier is used as quantitative measure of disease. The measure used to compute the dissimilarity between two images is the crucial component in this approach, and we evaluate four different image dissimilarity measures in the experiments.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Image Dissimilarity-Based Quantification of Lung Disease from CT

Sørensen¹,

Loog²,

Lo³

et al. 2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. In this paper, we propose to classify medical images using dissimilarities computed between collections of regions of interest. The images are mapped into a dissimilarity space using an image dissimilarity measure, and a standard vector space-based classifier is applied in this space. The classification output of this approach can be used in computer aided-diagnosis problems where the goal is to detect the presence of abnormal regions or to quantify the extent or severity of abnormalities in these regions. The proposed approach is applied to quantify chronic obstructive pulmonary disease in computed tomography (CT) images, achieving an area under the receiver operating characteristic curve of 0.817. This is significantly better compared to combining individual region classifications into an overall image classification, and compared to common computerized quantitative measures in pulmonary CT.

show abstract

Section: Image Dissimilarity Spacementioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Image Dissimilarity-Based Quantification of Lung Disease from CT

Sørensen¹,

Loog²,

Lo³

et al. 2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…However, while the advantage of considering the notion of similarity between datasets instead of between feature vectors has been recognized [2][3][4], attempts at formulating such measures have been mostly application dependent, often relying heavily on heuristics. A notable exception is a proposed universal normalized compression metric (NCM) based on Kolmogorov's notion of algorithmic complexity [5].…”

Section: The Hidden Modelsmentioning

confidence: 99%

Data smashing: uncovering lurking order in data

Chattopadhyay

Lipson

2014

J. R. Soc. Interface.

View full text Add to dashboard Cite

From automatic speech recognition to discovering unusual stars, underlying almost all automated discovery tasks is the ability to compare and contrast data streams with each other, to identify connections and spot outliers. Despite the prevalence of data, however, automated methods are not keeping pace. A key bottleneck is that most data comparison algorithms today rely on a human expert to specify what 'features' of the data are relevant for comparison. Here, we propose a new principle for estimating the similarity between the sources of arbitrary data streams, using neither domain knowledge nor learning. We demonstrate the application of this principle to the analysis of data from a number of real-world challenging problems, including the disambiguation of electro-encephalograph patterns pertaining to epileptic seizures, detection of anomalous cardiac activity from heart sound recordings and classification of astronomical objects from raw photometry. In all these cases and without access to any domain knowledge, we demonstrate performance on a par with the accuracy achieved by specialized algorithms and heuristics devised by domain experts. We suggest that data smashing principles may open the door to understanding increasingly complex observations, especially when experts do not know what to look for.

show abstract

“…, p r }, is needed. The dissimilarity representation allows to symbolize individual feature-patterns by pairwise dissimilarities computed between examples from the training set T and objects from the representation set R. Thus the dissimilarity vectors can be interpreted as numerical features and describe the relation between each object with the rest of objects [3].…”

Section: Introductionmentioning

confidence: 99%

One-Sided Prototype Selection on Class Imbalanced Dissimilarity Matrices

Millán-Giraldo

García

Sánchez

2012

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. In the dissimilarity representation paradigm, several prototype selection methods have been used to cope with the topic of how to select a small representation set for generating a low-dimensional dissimilarity space. In addition, these methods have also been used to reduce the size of the dissimilarity matrix. However, these approaches assume a relatively balanced class distribution, which is grossly violated in many real-life problems. Often, the ratios of prior probabilities between classes are extremely skewed. In this paper, we study the use of renowned prototype selection methods adapted to the case of learning from an imbalanced dissimilarity matrix. More specifically, we propose the use of these methods to under-sample the majority class in the dissimilarity space. The experimental results demonstrate that the one-sided selection strategy performs better than the classical prototype selection methods applied over all classes.

show abstract

Dissimilarity representations allow for building good classifiers

Cited by 193 publications

References 5 publications

Image Dissimilarity-Based Quantification of Lung Disease from CT

Image Dissimilarity-Based Quantification of Lung Disease from CT

Data smashing: uncovering lurking order in data

One-Sided Prototype Selection on Class Imbalanced Dissimilarity Matrices

Contact Info

Product

Resources

About