Ces dernières années, l'importance cruciale des métriques en apprentissage automatique a mené à un intérêt grandissant pour l'optimisation de distances et de similarités en utilisant l'information contenue dans des données d'apprentissage pour les rendre adaptées au problème traité. Ce domaine de recherche est souvent appelé apprentissage de métriques. En général, les méthodes existantes optimisent les paramètres d'une métrique devant respecter des contraintes locales sur les données d'apprentissage. Les métriques ainsi apprises sont généralement utilisées dans des algorithmes de plus proches voisins ou de clustering.Concernant les données numériques, beaucoup de travaux ont porté sur l'apprentissage de distance de Mahalanobis, paramétrisée par une matrice positive semi-définie. Les méthodes récentes sont capables de traiter des jeux de données de grande taille.Moins de travaux ont été dédiés à l'apprentissage de métriques pour les données structurées (comme les chaînes ou les arbres), car cela implique souvent des procédures plus complexes. La plupart des travaux portent sur l'optimisation d'une notion de distance d'édition, qui mesure (en termes de nombre d'opérations) le coût de transformer un objet en un autre.Au regard de l'état de l'art, nous avons identifié deux limites importantes des approches actuelles. Premièrement, elles permettent d'améliorer la performance d'algorithmes locaux comme les k plus proches voisins, mais l'apprentissage de métriques pour des algorithmes globaux (comme les classifieurs linéaires) n'a pour l'instant pas été beaucoup étudié. Le deuxième point, sans doute le plus important, est que la question de la capacité de généralisation des méthodes d'apprentissage de métriques a été largement ignorée.Dans cette thèse, nous proposons des contributions théoriques et algorithmiques qui répondent à ces limites. Notre première contribution est la construction d'un nouveau noyau construit à partir de probabilités d'édition apprises. A l'inverse d'autres noyaux entre chaînes, sa validité est garantie et il ne comporte aucun paramètre. Notre deuxième contribution est une nouvelle approche d'apprentissage de similarités d'édition pour les chaînes et les arbres inspirée par la théorie des (epsilon,gamma,tau)-bonnes fonctions de similarité et formulée comme un problème d'optimisation convexe. En utilisant la notion de stabilité uniforme, nous établissons des garanties théoriques pour la similarité apprise qui donne une borne sur l'erreur en généralisation d'un classifieur linéaire construit à partir de cette similarité. Dans notre troisième contribution, nous étendons ces principes à l'apprentissage de métriques pour les données numériques en proposant une méthode d'apprentissage de similarité bilinéaire qui optimise efficacement l'(epsilon,gamma,tau)-goodness. La similarité est apprise sous contraintes globales, plus appropriées à la classification linéaire. Nous dérivons des garanties théoriques pour notre approche, qui donnent de meilleurs bornes en généralisation pour le classifieur que dans le cas des données structurées. Notre dernière contribution est un cadre théorique permettant d'établir des bornes en généralisation pour de nombreuses méthodes existantes d'apprentissage de métriques. Ce cadre est basé sur la notion de robustesse algorithmique et permet la dérivation de bornes pour des fonctions de perte et des régulariseurs variés / In recent years, the crucial importance of metrics in machine learningalgorithms has led to an increasing interest in optimizing distanceand similarity functions using knowledge from training data to make them suitable for the problem at hand.This area of research is known as metric learning. Existing methods typically aim at optimizing the parameters of a given metric with respect to some local constraints over the training sample. The learned metrics are generally used in nearest-neighbor and clustering algorithms.When data consist of feature vectors, a large body of work has focused on learning a Mahalanobis distance, which is parameterized by a positive semi-definite matrix. Recent methods offer good scalability to large datasets.Less work has been devoted to metric learning from structured objects (such as strings or trees), because it often involves complex procedures. Most of the work has focused on optimizing a notion of edit distance, which measures (in terms of number of operations) the cost of turning an object into another.We identify two important limitations of current supervised metric learning approaches. First, they allow to improve the performance of local algorithms such as k-nearest neighbors, but metric learning for global algorithms (such as linear classifiers) has not really been studied so far. Second, and perhaps more importantly, the question of the generalization ability of metric learning methods has been largely ignored.In this thesis, we propose theoretical and algorithmic contributions that address these limitations. Our first contribution is the derivation of a new kernel function built from learned edit probabilities. Unlike other string kernels, it is guaranteed to be valid and parameter-free. Our second contribution is a novel framework for learning string and tree edit similarities inspired by the recent theory of (epsilon,gamma,tau)-good similarity functions and formulated as a convex optimization problem. Using uniform stability arguments, we establish theoretical guarantees for the learned similarity that give a bound on the generalization error of a linear classifier built from that similarity. In our third contribution, we extend the same ideas to metric learning from feature vectors by proposing a bilinear similarity learning method that efficiently optimizes the (epsilon,gamma,tau)-goodness. The similarity is learned based on global constraints that are more appropriate to linear classification. Generalization guarantees are derived for our approach, highlighting that our method minimizes a tighter bound on the generalization error of the classifier. Our last contribution is a framework for establishing generalization bounds for a large class of existing metric learning algorithms. It is based on a simple adaptation of the notion of algorithmic robustness and allows the derivation of bounds for various loss functions and regularizers.

We study the approximate answers to OLAP queries on data warehouses. We consider the relative answers to OLAP queries on a schema, as distributions with the L1 distance and approximate the answers without storing the entire data warehouse. We first introduce three specific methods: the uniform sampling, the measure-based sampling and the statistical model. We introduce also an edit distance between data warehouses with edit operations adapted for data warehouses. Then, in the OLAP data exchange, we study how to sample each source and combine the samples to approximate any OLAP query. We next consider a streaming context, where a data warehouse is built by streams of different sources. We show a lower bound on the size of the memory necessary to approximate queries. In this case, we approximate OLAP queries with a finite memory. We describe also a method to discover the statistical dependencies, a new notion we introduce. We are looking for them based on the decision tree. We apply the method to two data warehouses. The first one simulates the data of sensors, which provide weather parameters over time and location from different sources. The second one is the collection of RSS from the web sites on Internet.

In recent years, the crucial importance of metrics in machine learningalgorithms has led to an increasing interest in optimizing distanceand similarity functions using knowledge from training data to make them suitable for the problem at hand.This area of research is known as metric learning. Existing methods typically aim at optimizing the parameters of a given metric with respect to some local constraints over the training sample. The learned metrics are generally used in nearest-neighbor and clustering algorithms.When data consist of feature vectors, a large body of work has focused on learning a Mahalanobis distance, which is parameterized by a positive semi-definite matrix. Recent methods offer good scalability to large datasets.Less work has been devoted to metric learning from structured objects (such as strings or trees), because it often involves complex procedures. Most of the work has focused on optimizing a notion of edit distance, which measures (in terms of number of operations) the cost of turning an object into another.We identify two important limitations of current supervised metric learning approaches. First, they allow to improve the performance of local algorithms such as k-nearest neighbors, but metric learning for global algorithms (such as linear classifiers) has not really been studied so far. Second, and perhaps more importantly, the question of the generalization ability of metric learning methods has been largely ignored.In this thesis, we propose theoretical and algorithmic contributions that address these limitations. Our first contribution is the derivation of a new kernel function built from learned edit probabilities. Unlike other string kernels, it is guaranteed to be valid and parameter-free. Our second contribution is a novel framework for learning string and tree edit similarities inspired by the recent theory of (epsilon,gamma,tau)-good similarity functions and formulated as a convex optimization problem. Using uniform stability arguments, we establish theoretical guarantees for the learned similarity that give a bound on the generalization error of a linear classifier built from that similarity. In our third contribution, we extend the same ideas to metric learning from feature vectors by proposing a bilinear similarity learning method that efficiently optimizes the (epsilon,gamma,tau)-goodness. The similarity is learned based on global constraints that are more appropriate to linear classification. Generalization guarantees are derived for our approach, highlighting that our method minimizes a tighter bound on the generalization error of the classifier. Our last contribution is a framework for establishing generalization bounds for a large class of existing metric learning algorithms. It is based on a simple adaptation of the notion of algorithmic robustness and allows the derivation of bounds for various loss functions and regularizers.

We present a multifaceted investigation into the relevance of word order in machine translation. We introduce two tools, DTED and DERP, each using dependency structure to detect differences between the structures of machine-produced translations and human-produced references. DTED applies the principle of Tree Edit Distance to calculate edit operations required to convert one structure into another. Four variants of DTED have been produced, differing in the importance they place on words which match between the two sentences. DERP represents a more detailed procedure, making use of the dependency relations between words when evaluating the disparities between paths connecting matching nodes. In order to empirically evaluate DTED and DERP, and as a standalone contribution, we have produced WOJ-DB, a database of human judgments. Containing scores relating to translation adequacy and more specifically to word order quality, this is intended to support investigations into a wide range of translation phenomena. We report an internal evaluation of the information in WOJ-DB, then use it to evaluate variants of DTED and DERP, both to determine their relative merit and their strength relative to third-party baselines. We present our conclusions about the importance of structure to the tools and their relevance to word order specifically, then propose further related avenues of research suggested or enabled by our work.

The dissertation covers two problems motivated by computational biology: genome rearrangements, and haplotype networks.<p><p>Genome rearrangement problems are a particular case of edit distance problems, where one seeks to transform two given objects into one another using as few operations as possible, with the additional constraint that the set of allowed operations is fixed beforehand; we are also interested in computing the corresponding distances between those objects, i.e. merely computing the minimum number of operations rather than an optimal sequence. Genome rearrangement problems can often be formulated as sorting problems on permutations (viewed as linear orderings of {1,2,n}) using as few (allowed) operations as possible. In this thesis, we focus among other operations on ``transpositions', which displace intervals of a permutation. Many questions related to sorting by transpositions are open, related in particular to its computational complexity. We use the disjoint cycle decomposition of permutations, rather than the ``standard tools' used in genome rearrangements, to prove new upper bounds on the transposition distance, as well as formulae for computing the exact distance in polynomial time in many cases. This decomposition also allows us to solve a counting problem related to the ``cycle graph' of Bafna and Pevzner, and to construct a general framework for obtaining lower bounds on any edit distance between permutations by recasting their computation as factorisation problems on related even permutations.<p><p>Haplotype networks are graphs in which a subset of vertices is labelled, used in comparative genomics as an alternative to trees. We formalise a new method due to Cassens, Mardulyn and Milinkovitch, which consists in building a graph containing a given set of partially labelled trees and with as few edges as possible. We give exact algorithms for solving the problem on two graphs, with an exponential running time in the general case but with a polynomial running time if at least one of the graphs belong to a particular class.<p>/<p>La thèse couvre deux problèmes motivés par la biologie: l'étude des réarrangements génomiques, et celle des réseaux d'haplotypes.<p><p>Les problèmes de réarrangements génomiques sont un cas particulier des problèmes de distances d'édition, où l'on cherche à transformer un objet en un autre en utilisant le plus petit nombre possible d'opérations, les opérations autorisées étant fixées au préalable; on s'intéresse également à la distance entre les deux objets, c'est-à-dire au calcul du nombre d'opérations dans une séquence optimale plutôt qu'à la recherche d'une telle séquence. Les problèmes de réarrangements génomiques peuvent souvent s'exprimer comme des problèmes de tri de permutations (vues comme des arrangements linéaires de {1,2,n}) en utilisant le plus petit nombre d'opérations (autorisées) possible. Nous examinons en particulier les ``transpositions', qui déplacent un intervalle de la permutation. Beaucoup de problèmes liés au tri par transpositions sont ouverts, en particulier sa complexité algorithmique. Nous nous écartons des ``outils standards' utilisés dans le domaine des réarrangements génomiques, et utilisons la décomposition en cycles disjoints des permutations pour prouver de nouvelles majorations sur la distance des transpositions ainsi que des formules permettant de calculer cette distance en temps polynomial dans de nombreux cas. Cette décomposition nous sert également à résoudre un problème d'énumération concernant le ``graphe des cycles' de Bafna et Pevzner, et à construire une technique générale permettant d'obtenir de nouvelles minorations en reformulant tous les problèmes de distances d'édition sur les permutations en termes de factorisations de permutations paires associées.<p><p>Les réseaux d'haplotypes sont des graphes dont une partie des sommets porte des étiquettes, utilisés en génomique comparative quand les arbres sont trop restrictifs, ou quand l'on ne peut choisir une ``meilleure' topologie parmi un ensemble donné d'arbres. Nous formalisons une nouvelle méthode due à Cassens, Mardulyn et Milinkovitch, qui consiste à construire un graphe contenant tous les arbres partiellement étiquetés donnés et possédant le moins d'arêtes possible, et donnons des algorithmes résolvant le problème de manière optimale sur deux graphes, dont le temps d'exécution est exponentiel en général mais polynomial dans quelques cas que nous caractérisons.<p> / Doctorat en Sciences / info:eu-repo/semantics/nonPublished

This paper describes methods to track a user-defined point in the vision of a robot as it drives forward. This tracking allows a robot to keep itself directed at that point while driving so that it can get to that user-defined point. I develop and present two new multi-scale algorithms for tracking arbitrary points between two frames of video, as well as through a video sequence. The multi-scale algorithms do not use the traditional pyramid image, but instead use a data structure called an integral image (also known as a summed area table). The first algorithm uses edge-detection to track the movement of the tracking point between frames of video. The second algorithm uses a modified version of the Moravec operator to track the movement of the tracking point between frames of video. Both of these algorithms can track the user-specified point very quickly. Implemented on a conventional desktop, tracking can proceed at a rate of at least 20 frames per second.

Nous étudions les réponses proches à des requêtes OLAP sur les entrepôts de données. Nous considérons les réponses relatives aux requêtes OLAP sur un schéma, comme les distributions avec la distance L1 et rapprocher les réponses sans stocker totalement l'entrepôt de données. Nous présentons d'abord trois méthodes spécifiques: l'échantillonnage uniforme, l'échantillonnage basé sur la mesure et le modèle statistique. Nous introduisons également une distance d'édition entre les entrepôts de données avec des opérations d'édition adaptées aux entrepôts de données. Puis, dans l'échange de données OLAP, nous étudions comment échantillonner chaque source et combiner les échantillons pour rapprocher toutes requêtes OLAP. Nous examinons ensuite un contexte streaming, où un entrepôt de données est construit par les flux de différentes sources. Nous montrons une borne inférieure de la taille de la mémoire nécessaire aux requêtes approximatives. Dans ce cas, nous avons les réponses pour les requêtes OLAP avec une mémoire finie. Nous décrivons également une méthode pour découvrir les dépendances statistique, une nouvelle notion que nous introduisons. Nous recherchons ces dépendances en basant sur l'arbre de décision. Nous appliquons la méthode à deux entrepôts de données. Le premier simule les données de capteurs, qui fournissent des paramètres météorologiques au fil du temps et de l'emplacement à partir de différentes sources. Le deuxième est la collecte de RSS à partir des sites web sur Internet. / We study the approximate answers to OLAP queries on data warehouses. We consider the relative answers to OLAP queries on a schema, as distributions with the L1 distance and approximate the answers without storing the entire data warehouse. We first introduce three specific methods: the uniform sampling, the measure-based sampling and the statistical model. We introduce also an edit distance between data warehouses with edit operations adapted for data warehouses. Then, in the OLAP data exchange, we study how to sample each source and combine the samples to approximate any OLAP query. We next consider a streaming context, where a data warehouse is built by streams of different sources. We show a lower bound on the size of the memory necessary to approximate queries. In this case, we approximate OLAP queries with a finite memory. We describe also a method to discover the statistical dependencies, a new notion we introduce. We are looking for them based on the decision tree. We apply the method to two data warehouses. The first one simulates the data of sensors, which provide weather parameters over time and location from different sources. The second one is the collection of RSS from the web sites on Internet.

Les Graphes sont des structures mathématiques puissantes constituant un outil de modélisation universel utilisé dans différents domaines de l'informatique, notamment dans le domaine de la reconnaissance de formes. L'appariement de graphes est l'opération principale dans le processus de la reconnaissance de formes à base de graphes. Dans ce contexte, trouver des solutions d'appariement de graphes, garantissant l'optimalité en termes de précision et de temps de calcul est un problème de recherche difficile et d'actualité. Dans cette thèse, nous nous intéressons à la résolution de ce problème dans deux domaines : la reconnaissance de formes 2D et 3D. Premièrement, nous considérons le problème d'appariement de graphes géométriques et ses applications sur la reconnaissance de formes 2D. Dance cette première partie, la reconnaissance des Kites (structures archéologiques) est l'application principale considérée. Nous proposons un "framework" complet basé sur les graphes pour la reconnaissance des Kites dans des images satellites. Dans ce contexte, nous proposons deux contributions. La première est la proposition d'un processus automatique d'extraction et de transformation de Kites a partir d'images réelles en graphes et un processus de génération aléatoire de graphes de Kites synthétiques. En utilisant ces deux processus, nous avons généré un benchmark de graphes de Kites (réels et synthétiques) structuré en 3 niveaux de bruit. La deuxième contribution de cette première partie, est la proposition d'un nouvel algorithme d'appariement pour les graphes géométriques et par conséquent pour les Kites. L'approche proposée combine les invariants de graphes au calcul de l'édition de distance géométrique. Deuxièmement, nous considérons le problème de reconnaissance des formes 3D ou nous nous intéressons à la reconnaissance d'objets déformables représentés par des graphes c.à.d. des tessellations de triangles. Nous proposons une décomposition des tessellations de triangles en un ensemble de sous structures que nous appelons triangle-étoiles. En se basant sur cette décomposition, nous proposons un nouvel algorithme d'appariement de graphes pour mesurer la distance entre les tessellations de triangles. L'algorithme proposé assure un nombre minimum de structures disjointes, offre une meilleure mesure de similarité en couvrant un voisinage plus large et utilise un ensemble de descripteurs qui sont invariants ou au moins tolérants aux déformations les plus courantes. Finalement, nous proposons une approche plus générale de l'appariement de graphes. Cette approche est fondée sur une nouvelle formalisation basée sur le problème de mariage stable. L'approche proposée est optimale en terme de temps d'exécution, c.à.d. la complexité est quadratique O(n2), et flexible en terme d'applicabilité (2D et 3D). Cette approche se base sur une décomposition en sous structures suivie par un appariement de ces structures en utilisant l'algorithme de mariage stable. L'analyse de la complexité des algorithmes proposés et l'ensemble des expérimentations menées sur les bases de graphes des Kites (réelle et synthétique) et d'autres bases de données standards (2D et 3D) attestent l'efficacité, la haute performance et la précision des approches proposées et montrent qu'elles sont extensibles et générales / Graphs are powerful mathematical modeling tools used in various fields of computer science, in particular, in Pattern Recognition. Graph matching is the main operation in Pattern Recognition using graph-based approach. Finding solutions to the problem of graph matching that ensure optimality in terms of accuracy and time complexity is a difficult research challenge and a topical issue. In this thesis, we investigate the resolution of this problem in two fields: 2D and 3D Pattern Recognition. Firstly, we address the problem of geometric graphs matching and its applications on 2D Pattern Recognition. Kite (archaeological structures) recognition in satellite images is the main application considered in this first part. We present a complete graph based framework for Kite recognition on satellite images. We propose mainly two contributions. The first one is an automatic process transforming Kites from real images into graphs and a process of generating randomly synthetic Kite graphs. This allowing to construct a benchmark of Kite graphs (real and synthetic) structured in different level of deformations. The second contribution in this part, is the proposition of a new graph similarity measure adapted to geometric graphs and consequently for Kite graphs. The proposed approach combines graph invariants with a geometric graph edit distance computation. Secondly, we address the problem of deformable 3D objects recognition, represented by graphs, i.e., triangular tessellations. We propose a new decomposition of triangular tessellations into a set of substructures that we call triangle-stars. Based on this new decomposition, we propose a new algorithm of graph matching to measure the distance between triangular tessellations. The proposed algorithm offers a better measure by assuring a minimum number of triangle-stars covering a larger neighbourhood, and uses a set of descriptors which are invariant or at least oblivious under most common deformations. Finally, we propose a more general graph matching approach founded on a new formalization based on the stable marriage problem. The proposed approach is optimal in term of execution time, i.e. the time complexity is quadratic O(n2) and flexible in term of applicability (2D and 3D). The analyze of the time complexity of the proposed algorithms and the extensive experiments conducted on Kite graph data sets (real and synthetic) and standard data sets (2D and 3D) attest the effectiveness, the high performance and accuracy of the proposed approaches and show that the proposed approaches are extensible and quite general

A comparação de sequências finitas é uma ferramenta que é utilizada para a solução de problemas em várias áreas. Comparamos sequências inferindo quais são as operações de edição de substituição, inserção e remoção de símbolos que transformam uma sequência em uma outra. As matrizes de pontuação são estruturas largamente utilizadas e que definem um custo para cada tipo de operação de edição. Uma matriz de pontuação G é indexada pelos símbolos do alfabeto. A entrada de G na linha A, coluna B mede o custo da operação de edição para substituir o símbolo A pelo símbolo B. As matrizes de pontuação induzem funções que atribuem uma pontuação para um conjunto de operações de edição. Algumas dessas funções para a comparação de duas e de várias sequências são estudadas nesta tese. Quando cada símbolo de cada sequência é editado exatamente uma vez para transformar uma sequência em outra, o conjunto de operações de edição pode ser representado por uma estrutura conhecida por alinhamento. Descrevemos uma estrutura para representar o conjunto de operações de edição que não pode ser representado por um alinhamento convencional e descrevemos um algoritmo para encontrar a pontuação de uma sequência ótima de operações de edição usando um algoritmo conhecido para encontrar a pontuação de um alinhamento convencional ótimo. Considerando três diferentes funções induzidas de pontuação, caracterizamos, para cada uma delas, a classe das matrizes para as quais as funções induzidas de pontuação são métricas nas sequências. Dadas duas matrizes de pontuação G e G\', dizemos que elas são equivalentes para uma dada função que é induzida por uma matriz de pontuação e que avalia a qualidade de um alinhamento se, para quaisquer dois alinhamentos A e B, vale o seguinte: o alinhamento A é ``melhor\'\' do que o alinhamento B considerando a matriz G se e somente se A é ``melhor\'\' do que o alinhamento B considerando a matriz G\'. Neste trabalho, determinamos condições necessárias e suficientes para que duas matrizes de pontuação sejam equivalentes. Finalmente, definimos três novos critérios para pontuar alinhamentos de várias sequências. Todos os critérios consideram o comprimento do alinhamento além das operações de edição por ele representadas. Para cada um dos critérios definidos,propomos um algoritmo e o problema de decisão correspondente mostramos ser NP-completo. / Comparison of finite sequences is a tool used to solve problems in several areas. In order to compare sequences, we infer which are the edit operations of substitution, insertion and deletion of symbols that transform one sequence into another. Scoring matrices are a widely used structure to define a cost for each type of edit operation. A scoring matrix G is indexed by symbols of an alphabet. The entry in G in row A and column B measures the cost of the edit operation for replacing symbol A by symbol B. Scoring matrices induce functions that assign a score for a set of edit operations. Some of these functions for comparing two and multiple sequences are studied in this thesis. If each symbol is edited exactly once for transforming a sequence into another, the set of edit operations can be represented by a structure called alignment. We describe a structure to represent the set of edit operations that cannot be represented by a conventional alignment and we design an algorithm to find the cost of an optimal sequence of edit operations by using a known algorithm to find the cost of an optimal alignment. Considering three different kinds of induced scoring functions, we characterize, for each one of them, the class of matrices for which the induced scoring functions are metrics on sequences. Given two scoring matrices G and G\', we say they are equivalent for a given function that is induced by a scoring matrix and that evaluates the quality of an alignment if, for any two alignments A and B of two sequences, we have the following: alignment A is ``better\'\' than B considering scoring matrix G if and only if A is ``better\'\' than B considering scoring matrix G\'. In this work, we determine necessary and sufficient conditions for scoring matrices to be equivalent. Finally, we define three new criteria for scoring alignments of several sequence. Every criterion considers the length of the alignment and the edit operations represented by it. An algorithm for each criterion is studied and the corresponding decision problem is shown to be NP-complete.

En raison de la capacité et de l'amélioration des performances informatiques, les représentations structurelles sont devenues de plus en plus populaires dans le domaine de la reconnaissance de formes (RF). Quand les objets sont structurés à base de graphes, le problme de la comparaison d'objets revient à un problme d'appariement de graphes (Graph Matching). Au cours de la dernière décennie, les chercheurs travaillant dans le domaine de l'appariement de graphes ont porté une attention particulière à la distance d'édition entre graphes (GED), notamment pour sa capacité à traiter différent types de graphes. GED a été ainsi appliquée sur des problématiques spécifiques qui varient de la reconnaissance de molécules à la classi fication d'images. / Due to the inherent genericity of graph-based representations, and thanks to the improvement of computer capacities, structural representations have become more and more popular in the field of Pattern Recognition (PR). In a graph-based representation, vertices and their attributes describe objects (or part of them) while edges represent interrelationships between the objects. Representing objects by graphs turns the problem of object comparison into graph matching (GM) where correspondences between vertices and edges of two graphs have to be found.