We study sublinear-time algorithms for testing fundamental properties of functions and graphs. Sublinear-time algorithms run in time sublinear in the input size and are especially useful for processing massive datasets. We investigate variants of the property testing model, which was introduced by Rubinfeld and Sudan (SIAM J. Comput., 1996) and Goldreich, Goldwasser and Ron (J. ACM, 1998) to formally study sublinear-time algorithms. The focus of this thesis is on designing better property testers, deepening our understanding of their limitations, and proposing new angles for investigating sublinear-time algorithms. First, we study the problem of testing unateness of real-valued functions over high-dimensional domains. A function over a multi-dimensional domain is unate if, along each dimension, the function is either nonincreasing or nondecreasing. We give efficient unateness testers and prove that they are optimal. We then focus on testing monotonicity of functions when the input functions are partially corrupted or partially erased. We prove that, for some settings of parameters, testing monotonicity of functions with corrupted or erased values is exponentially harder than testing monotonicity of functions with neither corruptions nor erasures. We then investigate the parameters in terms of which the complexity of property testers should be expressed. The complexity of algorithms and, in particular, property testers is usually expressed in terms of the size of the input. We prove that certain parameters capture the complexity of testing problems better than the input size, providing compelling evidence that the input size may not always be the best parameter to express the complexity of sublinear-time algorithms. Finally, we turn our attention towards testing properties of graphs. We first provide an algorithm for testing connectedness of graphs and prove that it is optimal. We then define a new model for studying graphs with missing information and investigate the problems of testing connectedness and estimating the average degree of graphs in the new model.

We prove the existence and nonexistence of solutions for the semilinear problem ∆u + K(r)f(u) = 0 with various boundary conditions on the exterior of the ball in R^N such that lim r→∞u(r) = 0. Here f : R → R is an odd locally lipschitz non-linear function such that there exists a β > 0 with f < 0 on (0, β), f > 0 on (β, ∞), and K(r) \equiv r^−α for some α > 0.

No presente trabalho iremos provar, usando a folheação de Brouwer-Le Calvez e a teoria de forcing dela derivada, que dado um homeomorfismo f do toro isotópico à identidade tal que seu conjunto de rotação é um segmento de reta com inclinação irracional e tendo 0 como um ponto extremal, então f não possui difusão sublinear na direção perpendicular à direção do conjunto de rotação / In the present work we will prove, using the Brouwer-Le Calvez foliation and the forcing theory derived from it, that given a torus homeomorphism f isotopopic to the identity such that its rotation set is a line segment with irrational slope and 0 is an extreme point, then f does not have sublinear diffusion in the direction perpendicular to the direction of the rotation set.

This thesis initiates the study of the MAX-CUT problem in fully dynamic graphs. Given a graph $G=(V,E)$, we present the first fully dynamic algorithms to maintain a $\frac{1}{2}$-approximate cut in sublinear update time under edge insertions and deletions to $G$. Our results include the following deterministic algorithms: i) an $O(\Delta)$ \textit{worst-case} update time algorithm, where $\Delta$ denotes the maximum degree of $G$ and ii) an $O(m^{1/2})$ amortized update time algorithm where $m$ denotes the maximum number of edges in $G$ during any sequence of updates. \\ \indent We also give the following randomized algorithms when edge updates come from an oblivious adversary: i) a $\tilde{O}(n^{2/3})$ update time algorithm\footnote{Throughout this thesis, $\tilde{O}$ hides a $O(\text{polylog}(n))$ factor.} to maintain a $\frac{1}{2}$-approximate cut, and ii) a $\min\{\tilde{O}(n^{2/3}), \tilde{O}(\frac{n^{{3/2}+2c_0}}{m^{1/2}})\}$ worst case update time algorithm which maintains a $(\frac{1}{2}-o(1))$-approximate cut for any constant $c_0>0$ with high probability. The latter algorithm is obtained by designing a fully dynamic algorithm to maintain a sparse subgraph with sublinear (in $n$) maximum degree which approximates all large cuts in $G$ with high probability. / Graduate

Families of operators that are triangularizable must necessarily satisfy a number of spectral mapping properties. These necessary conditions are often sufficient as well. This thesis investigates such properties in finite dimensional and infinite dimensional Banach spaces. In addition, we investigate whether approximate spectral mapping conditions (being "close" in some sense) is similarly a sufficient condition.

Families of operators that are triangularizable must necessarily satisfy a number of spectral mapping properties. These necessary conditions are often sufficient as well. This thesis investigates such properties in finite dimensional and infinite dimensional Banach spaces. In addition, we investigate whether approximate spectral mapping conditions (being "close" in some sense) is similarly a sufficient condition.

<p>Collecting user data is crucial for advancing machine learning, social science, and government policies, but the privacy of the users whose data is being collected is a growing concern. {\em Differential Privacy (DP)} has emerged as the most standard notion for privacy protection with robust mathematical guarantees. Analyzing such massive amounts of data in a privacy-preserving manner motivates the need to study differentially-private algorithms that are also super-efficient.  </p> <p><br></p> <p>This thesis initiates a systematic study of differentially-private sublinear-time and sublinear-space algorithms. The contributions of this thesis are two-fold. First, we design some of the first differentially private sublinear algorithms for many fundamental problems. Second, we develop general DP techniques for designing differentially-private sublinear algorithms. </p> <p><br></p> <p>We give the first DP sublinear algorithm for clustering by generalizing a subsampling framework from the non-DP sublinear-time literature. We give the first DP sublinear algorithm for estimating the maximum matching size. Our DP sublinear algorithm for estimating the average degree of the graph achieves a better approximation than previous works. We give the first DP algorithm for releasing $L_2$-heavy hitters in the sliding window model and a pure $L_1$-heavy hitter algorithm in the same model, which improves upon previous works.  </p> <p><br></p> <p>We develop general techniques that address the challenges of designing sublinear DP algorithms. First, we introduce the concept of Coupled Global Sensitivity (CGS). Intuitively, the CGS of a randomized algorithm generalizes the classical  notion of global sensitivity of a function, by considering a coupling of the random coins of the algorithm when run on neighboring inputs. We show that one can achieve pure DP by adding Laplace noise proportional to the CGS of an algorithm. Second, we give a black box DP transformation for a specific class of approximation algorithms. We show that such algorithms can be made differentially private without sacrificing accuracy, as long as the function has small global sensitivity. In particular, this transformation gives rise to sublinear DP algorithms for many problems, including triangle counting, the weight of the minimum spanning tree, and norm estimation.</p>

No description available.

Dans cette thèse on considère deux modèles de calcul qui abordent des problèmes qui se posent lors du traitement des grosses données. Le premier modèle est le modèle de streaming. Lors du traitement des grosses données, un accès aux données de façon aléatoire est trop couteux. Les algorithmes de streaming ont un accès restreint aux données: ils lisent les données de façon séquentielle (par passage) une fois ou peu de fois. De plus, les algorithmes de streaming utilisent une mémoire d'accès aléatoire de taille sous-linéaire dans la taille des données. Le deuxième modèle est le modèle de communication. Lors du traitement des données par plusieurs entités de calcul situées à des endroits différents, l'échange des messages pour la synchronisation de leurs calculs est souvent un goulet d'étranglement. Il est donc préférable de minimiser la quantité de communication. Un modèle particulier est la communication à sens unique entre deux participants. Dans ce modèle, deux participants calculent un résultat en fonction des données qui sont partagées entre eux et la communication se réduit à un seul message. On étudie les problèmes suivants: 1) Les couplages dans le modèle de streaming. L'entrée du problème est un flux d'arêtes d'un graphe G=(V,E) avec n=|V|. On recherche un algorithme de streaming qui calcule un couplage de grande taille en utilisant une mémoire de taille O(n polylog n). L'algorithme glouton remplit ces contraintes et calcule un couplage de taille au moins 1/2 fois la taille d'un couplage maximum. Une question ouverte depuis longtemps demande si l'algorithme glouton est optimal si aucune hypothèse sur l'ordre des arêtes dans le flux est faite. Nous montrons qu'il y a un meilleur algorithme que l'algorithme glouton si les arêtes du graphe sont dans un ordre uniformément aléatoire. De plus, nous montrons qu'avec deux passages on peut calculer un couplage de taille strictement supérieur à 1/2 fois la taille d'un couplage maximum sans contraintes sur l'ordre des arêtes. 2) Les semi-couplages en streaming et en communication. Un semi-couplage dans un graphe biparti G=(A,B,E) est un sous-ensemble d'arêtes qui couple tous les sommets de type A exactement une fois aux sommets de type B de façon pas forcement injective. L'objectif est de minimiser le nombre de sommets de type A qui sont couplés aux même sommets de type B. Pour ce problème, nous montrons un algorithme qui, pour tout 0<=ε<=1, calcule une O(n^((1-ε)/2))-approximation en utilisant une mémoire de taille Ô(n^(1+ε)). De plus, nous montrons des bornes supérieures et des bornes inférieurs pour la complexité de communication entre deux participants pour ce problème et des nouveaux résultats concernant la structure des semi-couplages. 3) Validité des fichiers XML dans le modèle de streaming. Un fichier XML de taille n est une séquence de balises ouvrantes et fermantes. Une DTD est un ensemble de contraintes de validité locales d'un fichier XML. Nous étudions des algorithmes de streaming pour tester si un fichier XML satisfait les contraintes décrites dans une DTD. Notre résultat principal est un algorithme de streaming qui fait O(log n) passages, utilise 3 flux auxiliaires et une mémoire de taille O(log^2 n). De plus, pour le problème de validation des fichiers XML qui décrivent des arbres binaires, nous présentons des algorithmes en un passage et deux passages qui une mémoire de taille sous-linéaire. 4) Correction d'erreur pour la distance du cantonnier. Alice et Bob ont des ensembles de n points sur une grille en d dimensions. Alice envoit un échantillon de petite taille à Bob qui, après réception, déplace ses points pour que la distance du cantonnier entre les points d'Alice et les points de Bob diminue. Pour tout k>0 nous montrons qu'il y a un protocole presque optimal de communication avec coût de communication Ô(kd) tel que les déplacements des points effectués par Bob aboutissent à un facteur d'approximation de O(d) par rapport aux meilleurs déplacements de d points. / In this PhD thesis, we consider two computational models that address problems that arise when processing massive data sets. The first model is the Data Streaming Model. When processing massive data sets, random access to the input data is very costly. Therefore, streaming algorithms only have restricted access to the input data: They sequentially scan the input data once or only a few times. In addition, streaming algorithms use a random access memory of sublinear size in the length of the input. Sequential input access and sublinear memory are drastic limitations when designing algorithms. The major goal of this PhD thesis is to explore the limitations and the strengths of the streaming model. The second model is the Communication Model. When data is processed by multiple computational units at different locations, then the message exchange of the participating parties for synchronizing their calculations is often a bottleneck. The amount of communication should hence be as little as possible. A particular setting is the one-way two-party communication setting. Here, two parties collectively compute a function of the input data that is split among the two parties, and the whole message exchange reduces to a single message from one party to the other one. We study the following four problems in the context of streaming algorithms and one-way two-party communication: (1) Matchings in the Streaming Model. We are given a stream of edges of a graph G=(V,E) with n=|V|, and the goal is to design a streaming algorithm that computes a matching using a random access memory of size O(n polylog n). The Greedy matching algorithm fits into this setting and computes a matching of size at least 1/2 times the size of a maximum matching. A long standing open question is whether the Greedy algorithm is optimal if no assumption about the order of the input stream is made. We show that it is possible to improve on the Greedy algorithm if the input stream is in uniform random order. Furthermore, we show that with two passes an approximation ratio strictly larger than 1/2 can be obtained if no assumption on the order of the input stream is made. (2) Semi-matchings in Streaming and in Two-party Communication. A semi-matching in a bipartite graph G=(A,B,E) is a subset of edges that matches all A vertices exactly once to B vertices, not necessarily in an injective way. The goal is to minimize the maximal number of A vertices that are matched to the same B vertex. We show that for any 0<=ε<=1, there is a one-pass streaming algorithm that computes an O(n^((1-ε)/2))-approximation using Ô(n^(1+ε)) space. Furthermore, we provide upper and lower bounds on the two-party communication complexity of this problem, as well as new results on the structure of semi-matchings. (3) Validity of XML Documents in the Streaming Model. An XML document of length n is a sequence of opening and closing tags. A DTD is a set of local validity constraints of an XML document. We study streaming algorithms for checking whether an XML document fulfills the validity constraints of a given DTD. Our main result is an O(log n)-pass streaming algorithm with 3 auxiliary streams and O(log^2 n) space for this problem. Furthermore, we present one-pass and two-pass sublinear space streaming algorithms for checking validity of XML documents that encode binary trees. (4) Budget-Error-Correcting under Earth-Mover-Distance. We study the following one-way two-party communication problem. Alice and Bob have sets of n points on a d-dimensional grid [Δ]^d for an integer Δ. Alice sends a small sketch of her points to Bob and Bob adjusts his point set towards Alice's point set so that the Earth-Mover-Distance of Bob's points and Alice's points decreases. For any k>0, we show that there is an almost tight randomized protocol with communication cost Ô(kd) such that Bob's adjustments lead to an O(d)-approximation compared to the k best possible adjustments that Bob could make.

In this PhD thesis, we consider two computational models that address problems that arise when processing massive data sets. The first model is the Data Streaming Model. When processing massive data sets, random access to the input data is very costly. Therefore, streaming algorithms only have restricted access to the input data: They sequentially scan the input data once or only a few times. In addition, streaming algorithms use a random access memory of sublinear size in the length of the input. Sequential input access and sublinear memory are drastic limitations when designing algorithms. The major goal of this PhD thesis is to explore the limitations and the strengths of the streaming model. The second model is the Communication Model. When data is processed by multiple computational units at different locations, then the message exchange of the participating parties for synchronizing their calculations is often a bottleneck. The amount of communication should hence be as little as possible. A particular setting is the one-way two-party communication setting. Here, two parties collectively compute a function of the input data that is split among the two parties, and the whole message exchange reduces to a single message from one party to the other one. We study the following four problems in the context of streaming algorithms and one-way two-party communication: (1) Matchings in the Streaming Model. We are given a stream of edges of a graph G=(V,E) with n=|V|, and the goal is to design a streaming algorithm that computes a matching using a random access memory of size O(n polylog n). The Greedy matching algorithm fits into this setting and computes a matching of size at least 1/2 times the size of a maximum matching. A long standing open question is whether the Greedy algorithm is optimal if no assumption about the order of the input stream is made. We show that it is possible to improve on the Greedy algorithm if the input stream is in uniform random order. Furthermore, we show that with two passes an approximation ratio strictly larger than 1/2 can be obtained if no assumption on the order of the input stream is made. (2) Semi-matchings in Streaming and in Two-party Communication. A semi-matching in a bipartite graph G=(A,B,E) is a subset of edges that matches all A vertices exactly once to B vertices, not necessarily in an injective way. The goal is to minimize the maximal number of A vertices that are matched to the same B vertex. We show that for any 0<=ε<=1, there is a one-pass streaming algorithm that computes an O(n^((1-ε)/2))-approximation using Ô(n^(1+ε)) space. Furthermore, we provide upper and lower bounds on the two-party communication complexity of this problem, as well as new results on the structure of semi-matchings. (3) Validity of XML Documents in the Streaming Model. An XML document of length n is a sequence of opening and closing tags. A DTD is a set of local validity constraints of an XML document. We study streaming algorithms for checking whether an XML document fulfills the validity constraints of a given DTD. Our main result is an O(log n)-pass streaming algorithm with 3 auxiliary streams and O(log^2 n) space for this problem. Furthermore, we present one-pass and two-pass sublinear space streaming algorithms for checking validity of XML documents that encode binary trees. (4) Budget-Error-Correcting under Earth-Mover-Distance. We study the following one-way two-party communication problem. Alice and Bob have sets of n points on a d-dimensional grid [Δ]^d for an integer Δ. Alice sends a small sketch of her points to Bob and Bob adjusts his point set towards Alice's point set so that the Earth-Mover-Distance of Bob's points and Alice's points decreases. For any k>0, we show that there is an almost tight randomized protocol with communication cost Ô(kd) such that Bob's adjustments lead to an O(d)-approximation compared to the k best possible adjustments that Bob could make.