dissimilarity measure

Biological Mass Spectrometry is used to analyse peptides and proteins. A mass spectrum generates a list of measured mass to charge ratios and intensities of ionised peptides, which is called a peak-list. In order to classify the underlying amino acid sequence, the acquired spectra are usually compared with synthetic ones. Development of suitable methods of direct peak-list comparison may be advantageous for many applications. The pairwise peak-list comparison is a multistage process composed of matching of peaks embedded in two peak-lists, normalisation, scaling of peak intensities and dissimilarity measures. In our analysis, we focused on binary and intensity based measures. We have modified the measures in order to comprise the mass spectrometry specific properties of mass measurement accuracy and non-matching peaks. We compared the labelling of peak-list pairs, obtained using different factors of the pairwise peak-list comparison, as being the same or different to those determine...

In the field of Human-Computer Interaction (HCI), gesture recognition is becoming increasingly important as a mode of communication, in addition to the more common visual, aural and oral modes, and is of particular interest to designers of Augmentative and Alternative Communication (AAC) systems for people with disabilities. A complete microcomputer system is described, GesRec3D, which facilitates the data acquisition, segmentation, learning, and recognition of 3-Dimensional arm gestures. The gesture data is acquired from a Polhemus electro-magnetic tracker system, where sensors are placed on the finger, wrist and elbow of one arm. Coded gestures are linked to user-defined text, to be typed or spoken by a text-to-speech engine, which is integrated into the system. A segmentation method and an algorithm for classification are both presented, which includes acceptance/rejection thresholds based on intra-class and inter-class dissimilarity measures. Results of recognition hits, confusion misses and rejection misses are given for two experiments, involving predefined and arbitrary 3D gestures. Gestures, comprising complex shapes and movements of the body, are recognised and used effortlessly by humans, enabling a rich communication in combination with visual, aural and oral forms. However, whilst speech and image processing are mature areas in the field of Human-Computer Interaction (HCI), recognition of gestures is still relatively undeveloped. In addition to this, there is great interest in multi-modal HCI techniques that integrate speech, vision and gesture. Gesture recognition is also an important addition to Augmentative and Alternative Communication (AAC)

TOPS diagrams are concise descriptions of the structural topology of proteins, and their comparison usually relies on a structural alignment of the corresponding vertex ordered and vertex and edge labelled graphs. Such an approach involves checking for the existence of subgraph isomorphisms, which is an NP complete problem even for this kind of graphs. Therefore, although there exist several algorithms for the alignment-based comparison of TOPS diagrams that are fast in practice, they have an exponential worst case complexity. Moreover, the alignment-based comparison of TOPS diagrams assumes conservation of contiguity between homologous TOPS diagram segments. In this paper, we explore the alignment-free comparison of TOPS diagrams. We consider on the one hand similarity and dissimilarity measures based on subword composition of the sequences of secondary structure elements, thus neglecting contact map information, and on the other hand the Universal Similarity Metric from Kolmogorov complexity theory. Effectiveness of these alignment-free methods for TOPS diagrams comparison is assessed by cluster validation techniques.

Self-organized maps (SOM) have been applied to analyze the similarities of chemical compounds and to select from a given pool of descriptors the smallest and more relevant subset needed to build robust QSAR models based on fuzzy ARTMAP. First, the category maps for each molecular descriptor and for the target activity variable were created with SOM and then classified on the basis of topology and nonlinear distribution. The best subset of descriptors was obtained by choosing from each cluster the index with the highest correlation with the target variable and then in order of decreasing correlation. This process was terminated when a dissimilarity measure increased, indicating that the inclusion of more molecular indices would not add supplementary information. The optimal subset of descriptors was used as input to a fuzzy ARTMAP architecture modified to effect predictive capabilities. The performance of the integrated SOM-fuzzy ARTMAP approach was evaluated with the prediction of the acute toxicity LC 50 of a homogeneous set of 69 benzene derivatives in the fathead minnow and the oral rat toxicity LD 50 of a heterogeneous set of 155 organic compounds. The proposed methodology minimized the problem of misclassification of similar compounds and significantly enhanced the predictive capabilities of a properly trained fuzzy ARTMAP network.

Clustering is one of the most important data mining techniques that partitions data according to some similarity criterion. The problems of clustering categorical data have attracted much attention from the data mining research community recently. As the extension of the k-Means algorithm, the k-Modes algorithm has been widely applied to categorical data clustering by replacing means with modes. In this paper, the limitations of the simple matching dissimilarity measure and Ng's dissimilarity measure are analyzed using some illustrative examples. Based on the idea of biological and genetic taxonomy and rough membership function, a new dissimilarity measure for the k-Modes algorithm is defined. A distinct characteristic of the new dissimilarity measure is to take account of the distribution of attribute values on the whole universe. A convergence study and time complexity of the k-Modes algorithm based on new dissimilarity measure indicates that it can be effectively used for large data sets. The results of comparative experiments on synthetic data sets and five real data sets from UCI show the effectiveness of the new dissimilarity measure, especially on data sets with biological and genetic taxonomy information.

Clustering is one of the most important data mining techniques that partitions data according to some similarity criterion. The problems of clustering categorical data have attracted much attention from the data mining research community recently. As the extension of the k-Means algorithm, the k-Modes algorithm has been widely applied to categorical data clustering by replacing means with modes. In this paper, the limitations of the simple matching dissimilarity measure and Ng's dissimilarity measure are analyzed using some illustrative examples. Based on the idea of biological and genetic taxonomy and rough membership function, a new dissimilarity measure for the k-Modes algorithm is defined. A distinct characteristic of the new dissimilarity measure is to take account of the distribution of attribute values on the whole universe. A convergence study and time complexity of the k-Modes algorithm based on new dissimilarity measure indicates that it can be effectively used for large data sets. The results of comparative experiments on synthetic data sets and five real data sets from UCI show the effectiveness of the new dissimilarity measure, especially on data sets with biological and genetic taxonomy information.

Topographic maps such as the self organizing map (SOM) or neural gas (NG) constitute powerful data mining techniques which allow to simultaneously cluster data and infer its topological structure, such that additional features, e.g. browsing, become available. Both methods have been introduced for vectorial data sets, i.e. they require a classical feature encoding of information. Often, data are available in the form of pairwise distances only, such as e.g. arise from a kernel matrix, a graph, or some general dissimilarity measure. In such cases, NG and SOM cannot be applied directly. In this contribution, we introduce relational topographic maps as an extension of relational clustering algorithms which offer prototype-based representations of dissimilarity data, to incorporate neighborhood structure. These methods are equivalent to the standard (vectorial) techniques, if an Euclidean embedding exists, while preventing the necessity to explicitly compute such an embedding. Extending these techniques for the general case of non-Euclidean dissimilarities, an interpretation of relational clustering as clustering in pseudo-Euclidean space becomes possible. We compare the methods to wellknown clustering methods for proximity data based on deterministic annealing and discuss in how far convergence can be guaranteed in the general case. Relational clustering is quadratic in the number of data points which makes the algorithms infeasible for huge data sets. We propose an approximate patch-version of relational clustering which runs in linear time. The effectivity of the methods is demonstrated in a number of examples.

Topographic mapping offers a very flexible tool to inspect large quantities of high-dimensional data in an intuitive way. Often, electronic data are inherently non Euclidean and modern data formats are connected to dedicated non-Euclidean dissimilarity measures for which classical topographic mapping cannot be used. We give an overview about extensions of topographic mapping to general dissimilarities by means of median or relational extensions. Further, we discuss efficient approximations to avoid the usually squared time complexity.

Topographic maps such as the self-organizing map (SOM) or neural gas (NG) constitute powerful data mining techniques that allow simultaneously clustering data and inferring their topological structure, such that additional features, for example, browsing, become available. Both methods have been introduced for vectorial data sets; they require a classical feature encoding of information. Often data are available in the form of pairwise distances only, such as arise from a kernel matrix, a graph, or some general dissimilarity measure. In such cases, NG and SOM cannot be applied directly. In this article, we introduce relational topographic maps as an extension of relational clustering algorithms, which offer prototype-based representations of dissimilarity data, to incorporate neighborhood structure. These methods are equivalent to the standard (vectorial) techniques if a Euclidean embedding exists, while preventing the need to explicitly compute such an embedding. Extending these te...

This document is provided as an electronic appendix to the article "A new dissimilarity measure for finding semantic structure in category fluency data with implications for understanding memory organization in schizophrenia". This appendix contains two parts. In the first we derive a measure of the distance between two items in a fluency list that is appropriately normalized for list-length. This measure serves as the core element of the mean cumulative frequency (mcf) dissimilarity metric described in the article. In the second part we compare and contrast the mcf metric with the prevailing Chan et al. (1993) dissimilarity metric, referred to as the dis metric here and throughout the article. 1. Derivation of a Measure of Inter-item Distance based on Cumulative Frequencies Our measure of normalized inter-item distance is based on an analysis of the frequency distribution with which any given inter-item distance d occurs within lists of different lengths. In the following...

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.

A novel method for classification of abnormality in anatomical tree structures is presented. A tree is classified based on direct comparisons with other trees in a dissimilarity-based classification scheme. The pair-wise dissimilarity measure between two trees is based on a linear assignment between the branch feature vectors representing those trees. Hereby, localized information in the branches is collectively used in classification and variations in feature values across the tree are taken into account. An approximate anatomical correspondence between matched branches can be achieved by including anatomical features in the branch feature vectors. The proposed approach is applied to classify airway trees in computed tomography images of subjects with and without chronic obstructive pulmonary disease (COPD). Using the wall area percentage (WA%), a common measure of airway abnormality in COPD, as well as anatomical features to characterize each branch, an area under the receiver operating characteristic curve of 0.912 is achieved. This is significantly better than computing the average WA%.

This paper focuses on how to reduce cheating and minimize errors while automatically grading paper-based multiple-choice questions (MCQ) by making the whole process relatively fast, less expensive, more credible, and fairer especially when the number of examinees and number of questions are large. Credibility is obtained when techniques and best practices are introduced in the design process of MCQ. Fairness is obtained by personalizing evaluation through permutation of answers and questions. The distance introduced in personalization has led to the modification of the traditional automatic grading process where an application mapping the test number with its responses in the grading software is loaded automatically at each start of the grading process. On the extracted header fields, 2DFFT is applied as well as the reduction of computed coefficients to obtain the corresponding final local characteristic in the representation. The minimization of image processing errors is then obtained by training a support vector machine (SVM) for handwriting optical character recognition (OCR) using the Mixed National Institute of Standards and Technology (MNIST) dataset with 99.5% accuracy. The tests are carried out in several subjects at Fotso Victor University Institute of Technology (UIT) in Bandjoun and the ColTech of the University of Bamenda and teachers as well as students after investigation have confirmed that our method reduces cheating and improves the error rate during grading with fewer complaints.

Documents exist in different formats. When we have document images, in order to access some part, preferably all, of the information contained in that images, we have to deploy a document image analysis application. Document images can be mostly textual or mostly graphical. If, for a user, a task is to retrieve document images, relevant to a query from a set, we must use indexing techniques. The documents and the query are translated in a common representation. Using a dissimilarity measure (between the query and the document representations) and a method to speed-up the search process we may find documents that are from the user point of view relevant to his query. The semantic gap between a document representation and the user implicit representation can lead to unsatisfactory results. If we want to access objects from document images that are relevant to the document semantic we must enter in a document understanding cycle. Understanding document images is made in systems that are (usually) domain dependent, and that are not applicable in general cases (textual and graphical document classes). In this paper we present a method to describe and then to index document images using frequently occurences of items. The intuition is that frequent items represents symbols in a certain domain and this document description can be related to the domain knowledge (in an unsupervised manner). The novelty of our method consists in using graph summaries as a description for document images. In our approach we use a bag (multiset) of graphs as description for document images. From the document images we extract a graph based representation. In these graphs, we apply graph mining techniques in order to find frequent and maximally subgraphs. For each document image we construct a bag with all frequent subgraphs found in the graph-based representation. This bag of "symbols" represents the description of the document.

Dominant color descriptor (DCD) is one of the color descriptors proposed by MPEG-7 that has been extensively used for image retrieval. Among the color descriptors, DCD describes the salient color distributions in an image or a region of interest. DCD provides an effective, compact, and intuitive representation of colors presented in an image. In this paper, we will develop an efficient scheme for dominant color extraction. This approach significantly improves the efficiency of computation for dominant color extraction. In addition, we propose a modification for the MPEG-7 dissimilarity measure, which effectively improves the accuracy of perceptive similarity. Experimental results show that the proposed method achieves performance improvement not only in saving features extraction cost but also perceptually similar image retrieval.

This article presents the -distance, a family of distances between images recursively decomposed into segments and represented by multi-level feature vectors. Such a structure is a quad, a quin or a nona-tree resulting from a fixed and arbitrary image partition or from an image segmentation process. It handles positional information of image features (e.g. color, texture or shape). -distance is the generalized form of dissimilarity measures between multi-level feature vectors. Using different weights on tree nodes and different distances between nodes, distances between trees or visual similarity between images can be computed based on the general definition of . In this article, we present three -based distance families: two families of distances between tree structures, called T -distance (T for Tree) and S-distance (S for Segment), and a family of visual distances between images, called V-distance (V for Visual). The V-distance visually compares two images using their tree representation and the other two distances compare the tree structures resulting from image segmentation. Moreover, we show how existing distances between multi-level feature vectors appear to be particular cases of the -distance. Keywords Image database . Distance between quad/quin or nona-trees . Similarity of images . Similarity of image segments . Content-based image retrieval

This paper presents a tunable content-based music retrieval (CBMR) system suitable the for retrieval of music audio clips. The audio clips are represented as extracted feature vectors. The CBMR system is expert-tunable by altering the feature space. The feature space is tuned according to the expert-specified similarity criteria expressed in terms of clusters of similar audio clips. The main goal of tuning the feature space is to improve retrieval performance, since some features may have more impact on perceived similarity than others. The tuning process utilizes our genetic algorithm. The R-tree index for efficient retrieval of audio clips is based on the clustering of feature vectors. For each cluster a minimal bounding rectangle (MBR) is formed, thus providing objects for indexing. Inserting new nodes into the R-tree is efficiently performed because of the chosen Quadratic Split algorithm. Our CBMR system implements the point query and the n-nearest neighbors query with the O(log n) time complexity. Different objective functions based on cluster similarity and dissimilarity measures are used for the genetic algorithm. We have found that all of them have similar impact on the retrieval performance in terms of precision and recall. The paper includes experimental results in measuring retrieval performance, reporting significant improvement over the untuned feature space.

Relevance feedback (RF) is an iterative process, which refines the retrievals by utilizing the user's feedback on previously retrieved results. Traditional RF techniques solely use the short-term learning experience and do not exploit the knowledge created during cross sessions with multiple users. In this paper, we propose a novel RF framework, which facilitates the combination of shortterm and long-term learning processes by integrating the traditional methods with a new technique called the virtual feature. The feedback history with all the users is digested by the system and is represented in a very efficient form as a virtual feature of the images. As such, the dissimilarity measure can dynamically be adapted, depending on the estimate of the semantic relevance derived from the virtual features. In addition, with a dynamic database, the user's subject concepts may transit from one to another. By monitoring the changes in retrieval performance, the proposed system can automatically adapt the concepts according to the new subject concepts. The experiments are conducted on a real image database. The results manifest that the proposed framework outperforms the traditional within-session and log-based long-term RF techniques.

This work introduces a new family of link-based dissimilarity measures between nodes of a weighted, directed, graph that generalizes both the shortest-path and the commute-time (or resistance) distances. This measure, called the randomized shortest-path (RSP) dissimilarity, depends on a parameter θ and has the interesting property of reducing, on one end, to the standard shortest-path distance when θ is large and, on the other end, to the commute-time distance when θ is small (near zero). Intuitively, it corresponds to the expected cost incurred by a random walker in order to reach a destination node from a starting node, while maintaining a constant entropy (related to θ) spread in the graph. The parameter θ is therefore biasing gradually the simple random walk on the graph towards the shortest-path policy. By adopting a statistical physics approach and computing a sum over all the possible paths, it is shown that the RSP dissimilarity from every node to a particular node of interest can be computed efficiently by solving two linear systems of n equations, where n is the number of nodes. On the other hand, the dissimilarity between every couple of nodes is obtained by inverting an n × n matrix. The proposed measure could be used for various graph mining tasks such as computing betweenness centrality, finding dense communities, etc, as shown in the experimental section.

Advancing vulnerability science depends in part on identifying common themes from multiple, independent vulnerability assessments. Such insights are difficult to produce when the assessments use dissimilar, often qualitative, measures. The Vulnerability Scoping Diagram is presented to facilitate the comparison of assessments with dissimilar measures. The diagram is illustrated with recent research on drought vulnerabilities, showing that common insights into vulnerability may emerge if independent research teams use a common structure for organizing information about exposure, sensitivity and adaptive capacity-even if the underlying measures differ between assessments. Broadly adopting this technique, which is grounded in the ''Eight Steps '' methodological protocol [Schro¨ter, D., Polsky, C., Patt, A., 2005. Assessing vulnerabilities to the effects of global change: an eight step approach. Mitigation and Adaptation Strategies for Global Change 10(4), 573-595], will enable a vulnerability meta-analysis, the lessons from which may permit places to identify helpful adaptation or mitigation options without first having to conduct their own vulnerability assessments.

Many epidemiological studies involve analysis of clusters of diseases to infer locations of environmental hazards that could be responsible for the disease. This approach is however only suitable for sedentary populations or diseases with small latency periods. For migratory populations and diseases with long latency periods, people may change their residential location between time of exposure and onset of ill health. For such situations, clusters are diffused and diluted by in-and out-migration and may become very difficult to detect. One way to address the problem of diffused clusters is to include in analyses not only current residential locations, but all past locations at which cases might have been exposed to environmental hazardous. In this paper, we assume that a person's residential history provides such information and represent it through a discrete geospatial lifeline data model. Clusters of similar geospatial lifelines represent individuals who have similar residential histories-and therefore represent people who are more likely to have had similar environmental exposure histories. We therefore introduce a lifeline distance (dissimilarity) measure to detect clusters of cases, providing a basis for revealing possible regions in space-time where environmental hazards might have existed in the past. The ability of the measure to distinguish cases from controls is tested using two sets of synthetically generated cases and controls. Results indicate that the measure is able to consistently distinguish between populations of cases and controls with statistically significant results. The lifeline distance measure consistently outperforms another measure which uses only the distance between subjects' residences at time of diagnosis. However, the advantages of using the entire residential history are only partly realized, since the ability to distinguish between cases and controls is only moderately better for the lifeline distance function. Future work is needed to investigate This project is supported by grant number 1 R01 ES09816-01 from the National Institute of Environmental Health Sciences, NIH. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIEHS or NIH. We wish to thank Peter Rogerson for helpful discussions of the migration models, and the anonymous reviewers for pointing out areas where the paper could be improved.

This document is provided as an electronic appendix to the article "A new dissimilarity measure for finding semantic structure in category fluency data with implications for understanding memory organization in schizophrenia". This appendix contains two parts. In the first we derive a measure of the distance between two items in a fluency list that is appropriately normalized for list-length. This measure serves as the core element of the mean cumulative frequency (mcf) dissimilarity metric described in the article. In the second part we compare and contrast the mcf metric with the prevailing Chan et al. (1993) dissimilarity metric, referred to as the dis metric here and throughout the article. 1. Derivation of a Measure of Inter-item Distance based on Cumulative Frequencies Our measure of normalized inter-item distance is based on an analysis of the frequency distribution with which any given inter-item distance d occurs within lists of different lengths. In the following...

There are many procedures in the available literature to perform prediction in ungauged basins. Commonly, the Euclidean metric is used as a proxy of the hydrologic dissimilarity. Here we propose a procedure to find a metric on the basis of dissimilarity measures that are estimated from pairwise empirical copula densities of runoff. A metric is then defined in an transformed space of basin descriptors, whose parameterization is obtained with a variance reducing technique. A hydrologic model was run in an ungauged basin with sets of global parameters obtained from the k nearest neighboring donor basins using various metrics to take into account the uncertainty of its parameterization. Hydrologic model parameters were regionalized with a multiscale parameter regionalization technique whose transfer function parameters were found via calibration. The streamflow in an ungauged basin was found as an ensemble streamflow prediction to account for the uncertainties of the transfer function p...

Abstract- This paper presents a new approach to clustering fuzzy data, called Extensional Tree (ET) clustering algorithm by defining a dendrogram over fuzzy data and using a new metric between fuzzy numbers based on α-cuts. All the similar previous methods extended FCM to support fuzzy data. The present work is based on hierarchical clustering algorithm to cluster fuzzy data. In this novel approach a dendrogram is drawn over fuzzy or crisp data and then the desired clusters are extracted. Finally we compare this approach with some of the newly presented methods in the literature. The major advantage of ET is its fault tolerance against noisy samples. The overall experiments show prominence of our proposed method in comparison with other presented works.