No description available.

In nature animal species often exist in groups. We talk of insect swarms, flocks of birds, packs of lions, herds of wildebeest etc. These are characterised by individuals interacting by following their own rules, privy only to local information. Robotic swarms or simulations can be used explore such interactions. Mathematical formulations can be constructed that encode similar ideas and allow us to explore the emergent group behaviours. Some behaviours show characteristics reminiscent of the phenomena of criticality. A bird flock may show near instantaneous collective shifts in direction: velocity changes that appear to correlated over distances much larger individual separations. Here we examine swarm systems inspired by flocks of birds and the role played by criticality. The first system, Particle Swarm Optimisation (PSO), is shown to behave optimally when operating close to criticality. The presence of a critical point in the algorithm’s operation is shown to derive from the swarm’s properties as a random dynamical system. Empirical results demonstrate that the optimality lies on or near this point. A modified PSO algorithm is presented which uses measures of the swarm’s diversity as a feedback signal to adjust the behaviour of the swarm. This achieves a statistically balanced mixture of exploration and exploitation behaviours in the resultant swarm. The problems of stagnation and parameter tuning often encountered in PSO are automatically avoided. The second system, Swarm Chemistry, consists of heterogeneous particles combined with kinetic update rules. It is known that, depending upon the parametric configuration, numerous structures visually reminiscent of biological forms are found in this system. The parameter set discovered here results in a cell-division-like behaviour (in the sense of prokaryotic fission). Extensions to the swarm system produces a swarm that shows repeated cell division. As such, this model demonstrates a behaviour of interest to theories regarding the origin of life.

Matrix population models are discrete in both time and state-space, where a matrix with density-dependent entries is used to project a population vector of a stage-structured population from one time to the next. Such models are useful for modeling populations with discrete categorizations (e.g. developmental cycles, communities of multiple species, differing sizes, etc.). We present a general matrix model of two interacting populations where one (the resident) has a stable cycle, and we analyze when the other population (the invader) can successfully invade. Specifically, we study the local bifurcations of coexistence cycles as the resident cycle destabilizes, where a cycle of length 1 corresponds to an equilibrium. We make no assumptions on the types of interactions between the populations or on the population structure of the resident; we consider when the invader's projection matrix is primitive or imprimitive and 2x2. The simplest biological scenarios for such structures are an iteroparous invader and a two-stage semelparous invader. When the invader has a primitive projection matrix, coexistence cycles (of the same period as the resident cycle) bifurcate from the resident-cycle. When the invader has an imprimitive two-stage projection matrix, two types of coexistence cycles bifurcate from the resident-cycle: cycles of the same period and cycles of double the period. In both the primitive and imprimitive cases, we provide diagnostic quantities to determine the direction of bifurcation and the stability of the bifurcating cycles. Because we only perform a local stability analysis, the only successful invasion provided by our results is through stable coexistence cycles. As we show in some simple examples, however, the invader may persist when the coexistence cycles are unstable through competitive exclusion where the branch of bifurcating cycles connects to a branch of invader attractors and creates a multi-attractor scenario known as a strong Allee effect.

La performance sportive est influencée par de nombreux facteurs, lesquels s’influencent eux-mêmes réciproquement. La complexité de ces facteurs et de leurs relations ayant été négligée par les chercheurs, l’objet de la présente thèse était de rendre compte de cette complexité, à l’aide de méthodes empruntées à l’approche dynamique. Nous avons pu montrer que (a) les joueurs de football les plus experts construisent leur représentations du jeu en cours (les liens entre actions réalisées sur le terrain) avec des niveaux de complexité les plus élevés; (b) en aviron, une organisation motrice complexe, impliquant des interactions entre de nombreuses composantes, sous-tend la génération des mouvements de rame en cours; (c) le momentum psychologique en aviron se caractérise par des changements psychologiques et de performance qui s’inscrivent dans l’histoire de la performance; et (d) la performance excellente se développe à partir des interactions en cours entre les facteurs personnels et environnementaux couplés. Ces différents éclairages montrent l’intérêt d’une approche de la complexité pour comprendre les processus de performance. / The processes involved in human performance seem inherently complex and dynamic. For example, in order to “read the game”, a soccer player must integrate all the information from the ongoing movements and positions of team members, the opponents, the relative positions between them, where the ball is located, etc. Furthermore, an individual’s motor performance, which is particularly crucial in sports, depends on various simultaneous processes at different levels of the motor system: Cells, muscles, limbs, the brain, etc. In addition, individuals and teams do not perform in a void, but in achievement contexts, in which they strive for their goals, and their psychological states and performance may fluctuate as a function of many personal and environmental factors. For example, an athlete may enter a positive or negative spiral when perceiving that he or she is progressing or regressing in relation to the preferred goal or outcome (e.g., the victory). This perception of progress and regress, and the positive and negative psychological and behavioral (performance) changes accompanying this perception, are called positive and negative psychological momentum (PM; e.g., Gernigon, Briki, & Eykens, 2010). Positive and negative PM can emerge from one’s (or the opponent’s) mistakes, referee decisions, crowd behaviors, one’s psychological and physical state at a certain moment, and the interactions between these factors (Taylor & Demick, 1994). In addition, switching from performance on a relatively short time frame to a long-term process, individuals develop their abilities over multiple years, and hence over many practice or competition occasions. Ultimately, very few individuals develop world-class performance (e.g., winning Olympic medals), and their excellent abilities develop out of a combination of a variety of personal and environmental factors in interaction (e.g., motivation, coaching, family support, practice; Simonton, 1999). The current dissertation aims to capture complex dynamic performance-related processes, including the topics illustrated above. This means that we examine complexity at different levels (psychological, behavioral), time scales (from one training or competition session up to a career), as well as the interrelation between the processes across different levels and time scales.

This thesis proposes novel algorithms for using Gaussian processes for Discriminative pose estimation. We overcome the traditional limitations of Gaussian processes, their cubic training complexity and their uni-modal predictive distribution by assembling them in a mixture of experts formulation. Our First contribution shows that by creating a large number of Fixed size Gaussian process experts, we can build a model that is able to scale to large data sets and accurately learn the multi-modal and non- linear mapping between image features and the subject’s pose. We demonstrate that this model gives state of the art performance compared to other discriminative pose estimation techniques.We then extend the model to automatically learn the size and location of each expert. Gaussian processes are able to accurately model non-linear functional regression problems where the output is given as a function of the input. However, when an individual Gaussian process is trained on data which contains multi-modalities, or varying levels of ambiguity, the Gaussian process is unable to accurately model the data. We propose a novel algorithm for learning the size and location of each expert in our mixture of Gaussian processes model to ensure that the training data of each expert matches the assumptions of a Gaussian process. We show that this model is able to out perform our previous mixture of Gaussian processes model.Our final contribution is a dynamics framework for inferring a smooth sequence of pose estimates from a sequence of independent predictive distributions. Discriminative pose estimation infers the pose of each frame independently, leading to jittery tracking results. Our novel algorithm uses a model of human dynamics to infer a smooth path through a sequence of Gaussian mixture models as given by our mixture of Gaussian processes model. We show that our algorithm is able to smooth and correct some mis- takes made by the appearance model alone, and outperform a baseline linear dynamical system.

Recently dominant sets, a generalization of the notion of the maximal clique to edge-weighted graphs, have proven to be an effective tool for unsupervised learning and have found applications in different domains. Although, they were initially established using optimization and graph theory concepts, recent work has shown fascinating connections with evolutionary game theory, that leads to the clustering game framework. However, considering size of today's data sets, existing methods need to be modified in order to handle massive data. Hence, in this research work, first we address the limitations of the clustering game framework for large data sets theoretically. We propose a new important question for the clustering community ``How can a cluster of a subset of a dataset be a cluster of the entire dataset?''. We show that, this problem is a coNP-hard problem in a clustering game framework. Thus, we modify the definition of a cluster from a stable concept to a non-stable but optimal one (Nash equilibrium). By experiments we show that this relaxation does not change the qualities of the clusters practically. Following this alteration and the fact that equilibriums are generally compact subsets of vertices, we design an effective strategy to find equilibriums representing well distributed clusters. After finding such equilibriums, a linear game theoretic relation is proposed to assign vertices to the clusters and partition the graph. However, the method inherits a space complexity issue, that is the similarities between every pair of objects are required which proves practically intractable for large data sets. To overcome this limitation, after establishing necessary theoretical tools for a special type of graphs that we call vertex-repeated graphs, we propose the scalable clustering game framework. This approach divides a data set into disjoint tractable size chunks. Then, the exact clusters of the entire data are approximated by the clusters of the chunks. In fact, the exact equilibriums of the entire graph is approximated by the equilibriums of the subsets of the graph. We show theorems that enable significantly improved time complexity for the model. The applications include, but are not limited to, the maximum weight clique problem, large data clustering and image segmentation. Experiments have been done on random graphs and the DIMACS benchmark for the maximum weight clique problem and on magnetic resonance images (MRI) of the human brain consisting of about 4 million examples for large data clustering. Also, on the Berkeley Segmentation Dataset, the proposed method achieves results comparable to the state of the art, providing a parallel framework for image segmentation and without any training phase. The results show the effectiveness and efficiency of our approach. In another part of this research work, we generalize the clustering game method to cluster uncertain data where the similarities between the data points are not exactly known, that leads to the uncertain clustering game framework. Here, contrary to the ensemble clustering approaches, where the results of different similarity matrices are combined, we focus on the average utilities of an uncertain game. We show that the game theoretical solutions provide stable clusters even in the presence of severe uncertainties. In addition, based on this framework, we propose a novel concept in uncertain data clustering so that every subset of objects can have a ''cluster degree''. Extensive experiments on real world data sets, as well as on the Berkeley image segmentation dataset, confirm the performance of the proposed method. And finally, instead of dividing a graph into chunks to make the clustering scalable, we study the effect of the spectral sparsification method based on sampling by effective resistance on the clustering outputs. Through experimental and theoretical observations, we show that the clustering results obtained from sparsified graphs are very similar to the results of the original non-sparsified graphs. The rand index is always at about 0.9 to 0.99 in our experiments even when lots of sparsification is done.

L'étude des ensembles de mots complexité linéaire joue un rôle très important dans la théorie de combinatoire des mots et dans la théorie des systèmes dynamiques symboliques.Cette famille d'ensembles comprend les ensembles de facteurs : d'un mot Sturmien ou d'un mot d'Arnoux-Rauzy, d'un codage d'échange d'intervalle, d'un point fixe d'un morphisme primitif, etc.L'enjeu principal de cette thèse est l'étude de systèmes dynamiques minimales, définis de façon équivalente comme ensembles factoriels de mots uniformément récurrents.Comme résultat principal nous considérons une hiérarchie naturelle de systèmes minimal contenante les ensembles neutres, les tree sets et les ensembles spéculaires.De plus, on va relier ces systèmes au groupe libre en utilisant les mots de retours et les bases de sous-groupes d'indice fini.L'on étude aussi les systèmes symboliques dynamiques engendrés par les échanges d'intervalle et les involutions linéaires, ce qui nous permet d'obtenir des exemples et des interprétations géométriques des familles d'ensembles que définis dans notre hiérarchie.L'un des principal outil utilisé ici est l'étude des extensions possibles d'un mot dans un ensemble, ce qui nous permet de déterminer des propriétés telles que la complexité factorielle.Dans ce manuscrit, nous définissons le graphe d'extension, un graphe non orienté associé à chaque mot $w$ dans un ensemble $S$ qui décrit les extensions possibles de $w$ dans $S$ à gauche et à droite.Dans cette thèse, nous présentons plusieurs classes d'ensembles de mots définis par les formes possibles que les graphes d'extensions des éléments dans l'ensemble peuvent avoir.L'une des conditions les plus faibles que nous allons étudier est la condition de neutralité: un mot $w$ est neutre si le nombre de paires $(a,b)$ de lettres telles que $awb in S$ est égal au nombre de lettres $a$ tel que $aw in S$ plus le nombre de lettres $b$ tel que $wb in S$ moins 1.Un ensemble tel que chaque mot non vide satisfait la condition de neutralité est appelé un ensemble neutre.Une condition plus forte est la condition de l'arbre: un mot $w$ satisfait cette condition si son graphe d'extension est à la fois acyclique et connecté.Un ensemble est appelé un tree set si tout mot non vide satisfait cette condition.La famille de tree sets récurrents apparaît comme fermeture naturelle de deux familles d'ensembles très importants : les facteurs d'un mot d'Arnoux-Rauzy et les ensembles d'échange d'intervalle.Nous présentons également les ensembles spéculaires, une sous-famille remarquable de tree sets.Il s'agit également de sous-ensembles de groupes qui forment une généralisation naturelle des groupes libres.Ces ensembles de mots sont une généralisation abstraite des codages naturelles d'échanges d'intervalle et d'involutions linéaires.Pour chaque classe d'ensembles considéré dans cette thèse, nous montrons plusieurs résultats concernant les propriétés de fermeture (sous décodage maximale bifixe ou par rapport aux mots dérivés), la cardinalité des codes bifixes et les de mots de retour, la connexion entre mots de retour et bases du groupe libre, ainsi qu'entre les codes bifixes et les sous-groupes du groupe libre.Chacun de ces résultats est prouvé en utilisant les hypothèses les plus faibles possibles / Sets of words of linear complexity play an important role in combinatorics on words and symbolic dynamics.This family of sets includes set of factors of Sturmian and Arnoux-Rauzy words, interval exchange sets and primitive morphic sets, that is, sets of factors of fixed points of primitive morphisms.The leading issue of this thesis is the study of minimal dynamical systems, also defined equivalently as uniformly recurrent sets of words.As a main result, we consider a natural hierarchy of minimal systems containing neutral sets, tree sets and specular sets.Moreover, we connect the minimal systems to the free group using the notions of return words and basis of subroups of finite index.Symbolic dynamical systems arising from interval exchanges and linear involutions provide us geometrical examples of this kind of sets.One of the main tool used here is the study of possible extensions of a word in a set, that allows us to determine properties such as the factor complexity.In this manuscript we define the extension graph, an undirected graph associated to each word $w$ in a set $S$ which describes the possible extensions of $w$ in $S$ on the left and the right.In this thesis we present several classes of sets of words defined by the possible shapes that the graphs of elements in the set can have.One of the weakest condition that we will study is the neutrality condition: a word $w$ is neutral if the number of pairs $(a, b)$ of letters such that $awb in S$ is equal to the number of letters $a$ such that $aw in S$ plus the number of letters $b$ such that $wb in S$ minus 1.A set such that every nonempty word satisfies the neutrality condition is called a neutral set.A stronger condition is the tree condition: a word $w$ satisfies this condition if its extension graph is both acyclic and connected.A set is called a tree set if any nonempty word satisfies this condition.The family of recurrent tree sets appears as a the natural closure of two known families, namely the Arnoux-Rauzy sets and the interval exchange sets.We also introduce specular sets, a remarkable subfamily of the tree sets.These are subsets of groups which form a natural generalization of free groups.These sets of words are an abstract generalization of the natural codings of interval exchanges and of linear involutions.For each class of sets considered in this thesis, we prove several results concerning closure properties (under maximal bifix decoding or under taking derived words), cardinality of the bifix codes and set of return words in these sets, connection between return words and basis of the free groups, as well as between bifix codes and subgroup of the free group.Each of these results is proved under the weakest possible assumptions

Cette thèse se situe à l'interface entre la géométrie algébrique et les systèmes dynamiques. Le but est d'analyser la dynamique des automorphismes (ou, plus généralement, des transformations birationnelles) de variété compactes kaehleriennes avec première classe de Chern nulle, notamment des variétés hyperkaehleriennes. J'étudie l'existence de structures géométriques invariantes par la dynamique, en particulier fibrations et feuilletages, sous des hypothèses sur l'action en cohomologie de la transformation considérée / This thesis lies at the interface between algebraic geometry and dynamical systems. The goal is to analyse the dynamical behaviour of automorphisms (or, more generally, of birational transformations) of compact Kaehler manifolds having trivial first Chern class, in particular of hyperkaehler manifolds. I study the existence of geometric structures which are preserved by the dynamics, in particular fibrations and foliations, under some assumptions about the cohomological action of the transformation

This thesis describes the comparison of time-lapse field observations of suncups on alpine snow with numerical simulations. The simulations consist of solutions to a nonlinear partial differential equation which exhibits spontaneous pattern formation from a low amplitude, random initial surface. Both the field observations and the numerical solutions are found to saturate at a characteristic height and fluctuate chaotically with time. The timescale of these fluctuations is found to be instrumental in determining the full set of parameters for the numerical model such that it mimics the nonlinear dynamics of suncups. These parameters in turn are related to the change in albedo of the snow surface caused by the presence of suncups. This suggests the more general importance of dynamical behaviour in gaining an understanding of pattern formation phenomena. / Science, Faculty of / Physics and Astronomy, Department of / Graduate

No description available.