Les processus ponctuels spatiaux forment une branche de la statistique spatiale utilisée dans des domaines d'application variés (foresterie, géo-marketing, sismologie, épidémiologie...) et développée par de récents travaux théoriques. Nous nous intéressons principalement dans cette thèse à l'apport de la théorie des processus ponctuels spatiaux pour des problèmes de positionnement optimal, ainsi que pour la définition de nouveaux indices de concentration basés sur les distances en économétrie. Le problème de positionnement optimal s'écrit souvent comme un problème d'optimisation prenant en compte des données geo-référencées auxquelles peuvent être associées des caractéristiques. Pour prendre en compte l'aléa, nous considérons ces données issues d'un processus ponctuel spatial pour résoudre un problème de positionnement stochastique plus réaliste qu'un modèle déterministe. A travers l'étude du positionnement optimal d'une nouvelle caserne de pompiers dans la région toulousaine, nous développons une méthode de résolution stochastique permettant de juger de la variabilité de la solution optimale et de traiter des bases de données volumineuses. L'approche implémentée est validée par des premiers résultats théoriques sur le comportement asymptotique des solutions optimales empiriques. La convergence presque sure des solutions optimales empiriques de l'étude de cas précédente est obtenue dans un cadre i.i.d. en utilisant la théorie de Vapnik-Cervonenkis. Nous obtenons aussi la convergence presque sure des solutions optimales empiriques, dans un cadre plus général, pour un problème de positionnement dérivé du problème de transport de Monge-Kantorovich. Nous nous intéressons ensuite à des indices de concentration basés sur des distances en économétrie. Ces indices de concentration peuvent s'écrire comme des estimateurs de caractéristiques du second ordre de processus ponctuels marqués. Nous définissons ensuite un estimateur non-paramétrique d'une nouvelle caractéristique d'un processus ponctuel spatial marqué définissant ainsi un nouvel indice de concentration améliorant ceux déjà existants. Dans un cadre asymptotique avec fenêtre d'observation bornée, notre estimateur est asymptotiquement sans biais.

Disposant d'un jeu de données sur des infections nosocomiales, nous utilisons des techniques de vraisemblance empirique pour construire des bandes de confiance pour certaines quantité d'intérêt. Cette étude nous amène à renforcer les outils déjà existants afin qu'ils s'adaptent à notre cadre. Nous présentons dans une première partie les outils mathématiques issus de la littérature que nous utilisons dans ce travail de thèse. Nous les appliquons ensuite à diverses situations et donnons de nouvelles démonstrations lorsque cela est nécessaire. Nous conduisons aussi des simulations et obtenons des résultats concrets concernant notre jeu de données. Enfin, nous détaillons les algorithmes utilisés.

Disposant d’un jeu de données sur des infections nosocomiales, nous utilisons des techniques de vraisemblance empirique pour construire des bandes de confiance pour certaines quantité d’intérêt. Cette étude nous amène à renforcer les outils déjà existants afin qu’ils s’adaptent à notre cadre. Nous présentons dans une première partie les outils mathématiques issus de la littérature que nous utilisons dans ce travail de thèse. Nous les appliquons ensuite à diverses situations et donnons de nouvelles démonstrations lorsque cela est nécessaire. Nous conduisons aussi des simulations et obtenons des résultats concrets concernant notre jeu de données. Enfin, nous détaillons les algorithmes utilisés. / The starting point of this thesis is a data set of nosocomial infectionsin an intensive care unit of a French hostipal. We focused our attention onbuilding confidence bands for some parameters of interest using empiricallikelihood techniques. In order to do so, we had to adapt and develop somealready existing methods so that they fit our setup.We begin by giving a state of the art of the different theories we use.We then apply them to different setups and demonstrate new results whenneeded. Finally, we conduct simulations and describe our algorithms.

We prove that there exist three distinct, comprehensive classes of (formal) contexts with polynomially many concepts. Namely: contexts which are nowhere dense, of bounded breadth or highly convex. Already present in G. Birkhoff's classic monograph is the notion of breadth of a lattice; it equals the number of atoms of a largest boolean suborder. Even though it is natural to define the breadth of a context as being that of its concept lattice, this idea had not been exploited before. We do this and establish many equivalences. Amongst them, it is shown that the breadth of a context equals the size of its largest minimal generator, its largest contranominal-scale subcontext, as well as the Vapnik-Chervonenkis dimension of both its system of extents and of intents. The polynomiality of the aforementioned classes is proven via upper bounds (also known as majorants) for the number of maximal bipartite cliques in bipartite graphs. These are results obtained by various authors in the last decades. The fact that they yield statements about formal contexts is a reward for investigating how two established fields interact, specifically Formal Concept Analysis (FCA) and graph theory. We improve considerably the breadth bound. Such improvement is twofold: besides giving a much tighter expression, we prove that it limits the number of minimal generators. This is strictly more general than upper bounding the quantity of concepts. Indeed, it automatically implies a bound on these, as well as on the number of proper premises. A corollary is that this improved result is a bound for the number of implications in the canonical basis too. With respect to the quantity of concepts, this sharper majorant is shown to be best possible. Such fact is established by constructing contexts whose concept lattices exhibit exactly that many elements. These structures are termed, respectively, extremal contexts and extremal lattices. The usual procedure of taking the standard context allows one to work interchangeably with either one of these two extremal structures. Extremal lattices are equivalently defined as finite lattices which have as many elements as possible, under the condition that they obey two upper limits: one for its number of join-irreducibles, other for its breadth. Subsequently, these structures are characterized in two ways. Our first characterization is done using the lattice perspective. Initially, we construct extremal lattices by the iterated operation of finding smaller, extremal subsemilattices and duplicating their elements. Then, it is shown that every extremal lattice must be obtained through a recursive application of this construction principle. A byproduct of this contribution is that extremal lattices are always meet-distributive. Despite the fact that this approach is revealing, the vicinity of its findings contains unanswered combinatorial questions which are relevant. Most notably, the number of meet-irreducibles of extremal lattices escapes from control when this construction is conducted. Aiming to get a grip on the number of meet-irreducibles, we succeed at proving an alternative characterization of these structures. This second approach is based on implication logic, and exposes an interesting link between number of proper premises, pseudo-extents and concepts. A guiding idea in this scenario is to use implications to construct lattices. It turns out that constructing extremal structures with this method is simpler, in the sense that a recursive application of the construction principle is not needed. Moreover, we obtain with ease a general, explicit formula for the Whitney numbers of extremal lattices. This reveals that they are unimodal, too. Like the first, this second construction method is shown to be characteristic. A particular case of the construction is able to force - with precision - a high number of (in the sense of "exponentially many'') meet-irreducibles. Such occasional explosion of meet-irreducibles motivates a generalization of the notion of extremal lattices. This is done by means of considering a more refined partition of the class of all finite lattices. In this finer-grained setting, each extremal class consists of lattices with bounded breadth, number of join irreducibles and meet-irreducibles as well. The generalized problem of finding the maximum number of concepts reveals itself to be challenging. Instead of attempting to classify these structures completely, we pose questions inspired by Turán's seminal result in extremal combinatorics. Most prominently: do extremal lattices (in this more general sense) have the maximum permitted breadth? We show a general statement in this setting: for every choice of limits (breadth, number of join-irreducibles and meet-irreducibles), we produce some extremal lattice with the maximum permitted breadth. The tools which underpin all the intuitions in this scenario are hypergraphs and exact set covers. In a rather unexpected, but interesting turn of events, we obtain for free a simple and interesting theorem about the general existence of "rich'' subcontexts. Precisely: every context contains an object/attribute pair which, after removed, results in a context with at least half the original number of concepts.

We prove that there exist three distinct, comprehensive classes of (formal) contexts with polynomially many concepts. Namely: contexts which are nowhere dense, of bounded breadth or highly convex. Already present in G. Birkhoff's classic monograph is the notion of breadth of a lattice; it equals the number of atoms of a largest boolean suborder. Even though it is natural to define the breadth of a context as being that of its concept lattice, this idea had not been exploited before. We do this and establish many equivalences. Amongst them, it is shown that the breadth of a context equals the size of its largest minimal generator, its largest contranominal-scale subcontext, as well as the Vapnik-Chervonenkis dimension of both its system of extents and of intents. The polynomiality of the aforementioned classes is proven via upper bounds (also known as majorants) for the number of maximal bipartite cliques in bipartite graphs. These are results obtained by various authors in the last decades. The fact that they yield statements about formal contexts is a reward for investigating how two established fields interact, specifically Formal Concept Analysis (FCA) and graph theory. We improve considerably the breadth bound. Such improvement is twofold: besides giving a much tighter expression, we prove that it limits the number of minimal generators. This is strictly more general than upper bounding the quantity of concepts. Indeed, it automatically implies a bound on these, as well as on the number of proper premises. A corollary is that this improved result is a bound for the number of implications in the canonical basis too. With respect to the quantity of concepts, this sharper majorant is shown to be best possible. Such fact is established by constructing contexts whose concept lattices exhibit exactly that many elements. These structures are termed, respectively, extremal contexts and extremal lattices. The usual procedure of taking the standard context allows one to work interchangeably with either one of these two extremal structures. Extremal lattices are equivalently defined as finite lattices which have as many elements as possible, under the condition that they obey two upper limits: one for its number of join-irreducibles, other for its breadth. Subsequently, these structures are characterized in two ways. Our first characterization is done using the lattice perspective. Initially, we construct extremal lattices by the iterated operation of finding smaller, extremal subsemilattices and duplicating their elements. Then, it is shown that every extremal lattice must be obtained through a recursive application of this construction principle. A byproduct of this contribution is that extremal lattices are always meet-distributive. Despite the fact that this approach is revealing, the vicinity of its findings contains unanswered combinatorial questions which are relevant. Most notably, the number of meet-irreducibles of extremal lattices escapes from control when this construction is conducted. Aiming to get a grip on the number of meet-irreducibles, we succeed at proving an alternative characterization of these structures. This second approach is based on implication logic, and exposes an interesting link between number of proper premises, pseudo-extents and concepts. A guiding idea in this scenario is to use implications to construct lattices. It turns out that constructing extremal structures with this method is simpler, in the sense that a recursive application of the construction principle is not needed. Moreover, we obtain with ease a general, explicit formula for the Whitney numbers of extremal lattices. This reveals that they are unimodal, too. Like the first, this second construction method is shown to be characteristic. A particular case of the construction is able to force - with precision - a high number of (in the sense of "exponentially many'') meet-irreducibles. Such occasional explosion of meet-irreducibles motivates a generalization of the notion of extremal lattices. This is done by means of considering a more refined partition of the class of all finite lattices. In this finer-grained setting, each extremal class consists of lattices with bounded breadth, number of join irreducibles and meet-irreducibles as well. The generalized problem of finding the maximum number of concepts reveals itself to be challenging. Instead of attempting to classify these structures completely, we pose questions inspired by Turán's seminal result in extremal combinatorics. Most prominently: do extremal lattices (in this more general sense) have the maximum permitted breadth? We show a general statement in this setting: for every choice of limits (breadth, number of join-irreducibles and meet-irreducibles), we produce some extremal lattice with the maximum permitted breadth. The tools which underpin all the intuitions in this scenario are hypergraphs and exact set covers. In a rather unexpected, but interesting turn of events, we obtain for free a simple and interesting theorem about the general existence of "rich'' subcontexts. Precisely: every context contains an object/attribute pair which, after removed, results in a context with at least half the original number of concepts.