In the realm of network science, many topics can be abstracted as graph problems, such as routing, connectivity enhancement, resource/frequency allocation and so on. Though most of them are NP-hard to solve, heuristics as well as approximation algorithms are proposed to achieve reasonably good results. Accordingly, this dissertation studies graph related problems encountered in real applications. Two problems studied in this dissertation are derived from wireless network, two more problems studied are under scenarios of FIWI and optical network, one more problem is in Radio- Frequency Identification (RFID) domain and the last problem is inspired by satellite deployment.

The objective of most of relay nodes placement problems, is to place the fewest number of relay nodes in the deployment area so that the network, formed by the sensors and the relay nodes, is connected. Under the fixed budget scenario, the expense involved in procuring the minimum number of relay nodes to make the network connected, may exceed the budget. In this dissertation, we study a family of problems whose goal is to design a network with “maximal connectedness” or “minimal disconnectedness”, subject to a fixed budget constraint. Apart from “connectivity”, we also study relay node problem in which degree constraint is considered. The balance of reducing the degree of the network while maximizing communication forms the basis of our d-degree minimum arrangement(d-MA) problem. In this dissertation, we look at several approaches to solving the generalized d-MA problem where we embed a graph onto a subgraph of a given degree.

In recent years, considerable research has been conducted on optical and FIWI networks. Utilizing a recently proposed concept “candidate trees” in optical network, this dissertation studies counting problem on complete graphs. Closed form expressions are given for certain cases and a polynomial counting algorithm for general cases is also presented. Routing plays a major role in FiWi networks. Accordingly to a novel path length metric which emphasizes on “heaviest edge”, this dissertation proposes a polynomial algorithm on single path computation. NP-completeness proof as well as approximation algorithm are presented for multi-path routing.

Radio-frequency identification (RFID) technology is extensively used at present for identification and tracking of a multitude of objects. In many configurations, simultaneous activation of two readers may cause a “reader collision” when tags are present in the intersection of the sensing ranges of both readers. This dissertation ad- dresses slotted time access for Readers and tries to provide a collision-free scheduling scheme while minimizing total reading time.

Finally, this dissertation studies a monitoring problem on the surface of the earth for significant environmental, social/political and extreme events using satellites as sensors. It is assumed that the impact of a significant event spills into neighboring regions and there will be corresponding indicators. Careful deployment of sensors, utilizing “Identifying Codes”, can ensure that even though the number of deployed sensors is fewer than the number of regions, it may be possible to uniquely identify the region where the event has taken place.

Models using feature interactions have been applied successfully in many areas such as biomedical analysis, recommender systems. The popularity of using feature interactions mainly lies in (1) they are able to capture the nonlinearity of the data compared with linear effects and (2) they enjoy great interpretability. In this thesis, I propose a series of formulations using feature interactions for real world problems and develop efficient algorithms for solving them.

Specifically, I first propose to directly solve the non-convex formulation of the weak hierarchical Lasso which imposes weak hierarchy on individual features and interactions but can only be approximately solved by a convex relaxation in existing studies. I further propose to use the non-convex weak hierarchical Lasso formulation for hypothesis testing on the interaction features with hierarchical assumptions. Secondly, I propose a type of bi-linear models that take advantage of interactions of features for drug discovery problems where specific drug-drug pairs or drug-disease pairs are of interest. These models are learned by maximizing the number of positive data pairs that rank above the average score of unlabeled data pairs. Then I generalize the method to the case of using the top-ranked unlabeled data pairs for representative construction and derive an efficient algorithm for the extended formulation. Last but not least, motivated by a special form of bi-linear models, I propose a framework that enables simultaneously subgrouping data points and building specific models on the subgroups for learning on massive and heterogeneous datasets. Experiments on synthetic and real datasets are conducted to demonstrate the effectiveness or efficiency of the proposed methods.

Many applications require efficient data routing and dissemination in Delay Tolerant Networks (DTNs) in order to maximize the throughput of data in the network, such as providing healthcare to remote communities, and spreading related information in Mobile Social Networks (MSNs). In this thesis, the feasibility of using boats in the Amazon Delta Riverine region as data mule nodes is investigated and a robust data routing algorithm based on a fountain code approach is designed to ensure fast and timely data delivery considering unpredictable boat delays, break-downs, and high transmission failures. Then, the scenario of providing healthcare in Amazon Delta Region is extended to a general All-or-Nothing (Splittable) Multicommodity Flow (ANF) problem and a polynomial time constant approximation algorithm is designed for the maximum throughput routing problem based on a randomized rounding scheme with applications to DTNs. In an MSN, message content is closely related to users’ preferences, and can be used to significantly impact the performance of data dissemination. An interest- and content-based algorithm is developed where the contents of the messages, along with the network structural information are taken into consideration when making message relay decisions in order to maximize data throughput in an MSN. Extensive experiments show the effectiveness of the above proposed data dissemination algorithm by comparing it with state-of-the-art techniques.

Many forms of programmable matter have been proposed for various tasks. We use an abstract model of self-organizing particle systems for programmable matter which could be used for a variety of applications, including smart paint and coating materials for engineering or programmable cells for medical uses. Previous research using this model has focused on shape formation and other spatial configuration problems, including line formation, compression, and coating. In this work we study foundational computational tasks that exceed the capabilities of the individual constant memory particles described by the model. These tasks represent new ways to use these self-organizing systems, which, in conjunction with previous shape and configuration work, make the systems useful for a wider variety of tasks. We present an implementation of a counter using a line of particles, which makes it possible for the line of particles to count to and store values much larger than their individual capacities. We then present an algorithm that takes a matrix and a vector as input and then sets up and uses a rectangular block of particles to compute the matrix-vector multiplication. This setup also utilizes the counter implementation to store the resulting vector from the matrix-vector multiplication. Operations such as counting and matrix multiplication can leverage the distributed and dynamic nature of the self-organizing system to be more efficient and adaptable than on traditional linear computing hardware. Such computational tools also give the systems more power to make complex decisions when adapting to new situations or to analyze the data they collect, reducing reliance on a central controller for setup and output processing. Finally, we demonstrate an application of similar types of computations with self-organizing systems to image processing, with an implementation of an image edge detection algorithm.

The power and communication networks are highly interdependent and form a part of the critical infrastructure of a country. Similarly, dependencies exist within the networks itself. Owing to cascading failures, interdependent and intradependent networks are extremely susceptible to widespread vulnerabilities. In recent times the research community has shown significant interest in modeling to capture these dependencies. However, many of them are simplistic in nature which limits their applicability to real world systems. This dissertation presents a Boolean logic based model termed as Implicative Interdependency Model (IIM) to capture the complex dependencies and cascading failures resulting from an initial failure of one or more entities of either network.

Utilizing the IIM, four pertinent problems encompassing vulnerability and protection of critical infrastructures are formulated and solved. For protection analysis, the Entity Hardening Problem, Targeted Entity Hardening Problem and Auxiliary Entity Allocation Problem are formulated. Qualitatively, under a resource budget, the problems maximize the number of entities protected from failure from an initial failure of a set of entities. Additionally, the model is also used to come up with a metric to analyze the Robustness of critical infrastructure systems. The computational complexity of all these problems is NP-complete. Accordingly, Integer Linear Program solutions (to obtain the optimal solution) and polynomial time sub-optimal Heuristic solutions are proposed for these problems. To analyze the efficacy of the Heuristic solution, comparative studies are performed on real-world and test system data.

Major Depression, clinically called Major Depressive Disorder, is a mood disorder that affects about one eighth of population in US and is projected to be the second leading cause of disability in the world by the year 2020. Recent advances in biotechnology have enabled us to collect a great variety of data which could potentially offer us a deeper understanding of the disorder as well as advancing personalized medicine.

This dissertation focuses on developing methods for three different aspects of predictive analytics related to the disorder: automatic diagnosis, prognosis, and prediction of long-term treatment outcome. The data used for each task have their specific characteristics and demonstrate unique problems. Automatic diagnosis of melancholic depression is made on the basis of metabolic profiles and micro-array gene expression profiles where the presence of missing values and strong empirical correlation between the variables is not unusual. To deal with these problems, a method of generating a representative set of features is proposed. Prognosis is made on data collected from rating scales and questionnaires which consist mainly of categorical and ordinal variables and thus favor decision tree based predictive models. Decision tree models are known for the notorious problem of overfitting. A decision tree pruning method that overcomes the shortcomings of a greedy nature and reliance on heuristics inherent in traditional decision tree pruning approaches is proposed. The method is further extended to prune Gradient Boosting Decision Tree and tested on the task of prognosis of treatment outcome. Follow-up studies evaluating the long-term effect of the treatments on patients usually measure patients' depressive symptom severity monthly, resulting in the actual time of relapse upper bounded by the observed time of relapse. To resolve such uncertainty in response, a general loss function where the hypothesis could take different forms is proposed to predict the risk of relapse in situations where only an interval for time of relapse can be derived from the observed data.

Imagine that we have a piece of matter that can change its physical properties like its shape, density, conductivity, or color in a programmable fashion based on either user input or autonomous sensing. This is the vision behind what is commonly known as programmable matter. Envisioning systems of nano-sensors devices, programmable matter consists of systems of simple computational elements, called particles, that can establish and release bonds, compute, and can actively move in a self-organized way. In this dissertation the feasibility of solving fundamental problems relevant for programmable matter is investigated. As a model for such self-organizing particle systems (SOPS), the geometric amoebot model is introduced. In this model, particles only have local information and have modest computational power. They achieve locomotion by expanding and contracting, which resembles the behavior of amoeba. Under this model, efficient local-control algorithms for the leader election problem in SOPS are presented. As a central problem for programmable matter, shape formation problems are then studied. The limitations of solving the leader election problem and the shape formation problem on a more general version of the amoebot model are also discussed. The \smart paint" problem is also studied which aims at having the particles self-organize in order to uniformly coat the surface of an object of arbitrary shape and size, forming multiple coating layers if necessary. A Universal Coating algorithm is presented and shown to be asymptotically worst-case optimal both in terms of time with high probability and work. In particular, the algorithm always terminates within a linear number of rounds with high probability. A linear lower bound on the competitive gap between fully local coating algorithms and coating algorithms that rely on global information is presented, which implies that the proposed algorithm is also optimal in a competitive sense. Simulation results show that the competitive ratio of the proposed algorithm may be better than linear in practice. Developed algorithms utilize only local control, require only constant-size memory particles, and are asymptotically optimal in terms of the total number of particle movements needed to reach the desired shape configuration.

Large-scale $\ell_1$-regularized loss minimization problems arise in high-dimensional applications such as compressed sensing and high-dimensional supervised learning, including classification and regression problems. In many applications, it remains challenging to apply the sparse learning model to large-scale problems that have massive data samples with high-dimensional features. One popular and promising strategy is to scaling up the optimization problem in parallel. Parallel solvers run multiple cores on a shared memory system or a distributed environment to speed up the computation, while the practical usage is limited by the huge dimension in the feature space and synchronization problems.

In this dissertation, I carry out the research along the direction with particular focuses on scaling up the optimization of sparse learning for supervised and unsupervised learning problems. For the supervised learning, I firstly propose an asynchronous parallel solver to optimize the large-scale sparse learning model in a multithreading environment. Moreover, I propose a distributed framework to conduct the learning process when the dataset is distributed stored among different machines. Then the proposed model is further extended to the studies of risk genetic factors for Alzheimer's Disease (AD) among different research institutions, integrating a group feature selection framework to rank the top risk SNPs for AD. For the unsupervised learning problem, I propose a highly efficient solver, termed Stochastic Coordinate Coding (SCC), scaling up the optimization of dictionary learning and sparse coding problems. The common issue for the medical imaging research is that the longitudinal features of patients among different time points are beneficial to study together. To further improve the dictionary learning model, I propose a multi-task dictionary learning method, learning the different task simultaneously and utilizing shared and individual dictionary to encode both consistent and changing imaging features.

Imaging genetics is an emerging and promising technique that investigates how genetic variations affect brain development, structure, and function. By exploiting disorder-related neuroimaging phenotypes, this class of studies provides a novel direction to reveal and understand the complex genetic mechanisms. Oftentimes, imaging genetics studies are challenging due to the relatively small number of subjects but extremely high-dimensionality of both imaging data and genomic data. In this dissertation, I carry on my research on imaging genetics with particular focuses on two tasks---building predictive models between neuroimaging data and genomic data, and identifying disorder-related genetic risk factors through image-based biomarkers. To this end, I consider a suite of structured sparse methods---that can produce interpretable models and are robust to overfitting---for imaging genetics. With carefully-designed sparse-inducing regularizers, different biological priors are incorporated into learning models. More specifically, in the Allen brain image--gene expression study, I adopt an advanced sparse coding approach for image feature extraction and employ a multi-task learning approach for multi-class annotation. Moreover, I propose a label structured-based two-stage learning framework, which utilizes the hierarchical structure among labels, for multi-label annotation. In the Alzheimer's disease neuroimaging initiative (ADNI) imaging genetics study, I employ Lasso together with EDPP (enhanced dual polytope projections) screening rules to fast identify Alzheimer's disease risk SNPs. I also adopt the tree-structured group Lasso with MLFre (multi-layer feature reduction) screening rules to incorporate linkage disequilibrium information into modeling. Moreover, I propose a novel absolute fused Lasso model for ADNI imaging genetics. This method utilizes SNP spatial structure and is robust to the choice of reference alleles of genotype coding. In addition, I propose a two-level structured sparse model that incorporates gene-level networks through a graph penalty into SNP-level model construction. Lastly, I explore a convolutional neural network approach for accurate predicting Alzheimer's disease related imaging phenotypes. Experimental results on real-world imaging genetics applications demonstrate the efficiency and effectiveness of the proposed structured sparse methods.

Passwords are ubiquitous and are poised to stay that way due to their relative usability, security and deployability when compared with alternative authentication schemes. Unfortunately, humans struggle with some of the assumptions or requirements that are necessary for truly strong passwords. As administrators try to push users towards password complexity and diversity, users still end up using predictable mangling patterns on old passwords and reusing the same passwords across services; users even inadvertently converge on the same patterns to a surprising degree, making an attacker’s job easier. This work explores using machine learning techniques to pick out strong passwords from weak ones, from a dataset of 10 million passwords, based on how structurally similar they were to the rest of the set.