Based on the author's D. Phil. Thesis (University of Oxford, 2003). / Includes bibliographical references. Available on-line.

Thesis (Ph. D.)--University of Texas at Austin, 2009. / Title from PDF title page (University of Texas Digital Repository, viewed on Sept. 16, 2009). Vita. Includes bibliographical references.

Application-layer overlay networks have recently emerged as a promising platform to deploy additional services over the Internet. A virtual network of overlay nodes can be used to regulate traffic flows of an underlay network, without modifying the underlay network infrastructure. As a result, an opportunity to redeem the inefficiency of IP routing and to improve end-to-end performance of the Internet has arisen, by routing traffic over multiple overlay paths. However, to achieve high end-to-end performance over the Internet by means of overlay networks, a number of challenging issues, including limited knowledge of the underlay network characteristics, fluctuations of overlay path performance, and interactions between overlay and the underlay traffic must be addressed. This thesis provides solutions to some of these issues, by proposing a framework to construct a multi-path overlay architecture for improving Internet end-to-end performance and availability. The framework is formed by posing a series of questions, including i) how to model and forecast overlay path performance characteristics; ii) how to route traffic optimally over multiple overlay paths; and iii) how to place overlay nodes to maximally leverage the Internet resource redundancy, while minimizing the deployment cost. To answer those research questions, analytical and experimental studies have been conducted. As a result, i) a loss model and a hybrid forecasting technique are proposed to capture, and subsequently predict end-to-end loss/delay behaviors; with this predictive capability, overlay agents can, for example, select overlay paths that potentially offer good performance and reliability; ii) to take full advantage of the predictive capability and the availability of multiple paths, a Markov Decision Process based multi-path traffic controller is developed, which can route traffic simultaneously over multiple overlay paths to optimize some performance measures, e.g. average loss rate and latency. As there can be multiple overlay controllers, competing for common resources by making selfish decisions, which could jeopardize performance of the networks, game theory is applied here to turn the competition into cooperation; as a consequence, the network performance is improved; iii) furthermore, to facilitate the deployment of the multi-path overlay architecture, a multi-objective genetic-based algorithm is introduced to place overlay nodes to attain a high level of overlay path diversity, while minimizing the number of overlay nodes to be deployed, and thus reducing the deployment cost. The findings of this thesis indicate that the use of multiple overlay paths can substantially improve end-to-end performance. They uncover the potential of multi-path application-layer overlay networks as an architecture for achieving high end-to-end performance and availability over the Internet.

Algorithms that use relational data are increasingly used to allocate resources within society. As researchers and decision-makers have adapted the role of algorithms from a descriptive one (describing patterns in data) to a prescriptive one (making decisions in predictive systems), there is an increasing concern that algorithms may replicate and even amplify societal bias, allocating worse or less resources to minorities and underrepresented groups. This dissertation proposes methodology for diagnosing when and how algorithms amplify inequality on networks as well as designing interventions for mitigating algorithmic bias. We leverage methods from network modeling, algorithmic game theory, and fair machine learning to uncover the root driver of bias in network data and to leverage this knowledge in order to design fair algorithms. In this thesis, we mostly focus on unsupervised learning problems, which present unique challenges that require a multi-faceted approach. We propose a unifying formulation for unifying different problems in unsupervised learning on networks and use it to propose methods to find the root cause of bias through modeling patterns of connections and embeddings. We leverage this knowledge to design fairer algorithms as well as to define diagnoses metrics for evaluating inequality before and after an algorithm is introduced. Furthermore, we argue for the need to bridge optimization-based learning and utility-based learning in creating stable, efficient, and useful systems. We use network models and mathematical formulations of distributional inequality in diagnosing the algorithmic amplification of bias in social recommendations and ranking algorithms. We find that the most common and neutral algorithms may further underrepresent minority groups in creating new connections or achieving high levels of visibility in networks that exhibit competition in increasing social capital and homophily (the tendency of people to connect with those similar to them). We uncover the role of homophily in helping a minority group overcome their initial disadvantage and we leverage it to design fairer information campaigns that equitable distribute messages across a population. Akin to this goal, we incorporate notions of utility and welfare in our algorithmic design, re-designing heuristics for grouping and clustering that improve the diversity of groups while preserving their usefulness, with applications in political and educational districting. Overall, this set of results aims to investigate the impact of algorithms on the outcomes of different populations and to open new avenues for inter-disciplinary research methods that can alleviate algorithmic bias. We close by discussing connections between different fields and methods as well as directions for future research.

Finding an optimal solution for the logic circuit design problem is challenging and time-consuming especially for complex logic circuits. As the number of logic gates increases the task of designing optimal logic circuits extends beyond human capability. A number of evolutionary algorithms have been invented to tackle a range of optimisation problems, including logic circuit design. This dissertation explores two of these evolutionary algorithms i.e. Gene Expression Programming (GEP) and Multi Expression Programming (MEP) with the aim of integrating their strengths into a new Genetic Programming (GP) algorithm. GEP was invented by Candida Ferreira in 1999 and published in 2001 [8]. The GEP algorithm inherits the advantages of the Genetic Algorithm (GA) and GP, and it uses a simple encoding method to solve complex problems [6, 32]. While GEP emerged as powerful due to its simplicity in implementation and exibility in genetic operations, it is not without weaknesses. Some of these inherent weaknesses are discussed in [1, 6, 21]. Like GEP, MEP is a GP-variant that uses linear chromosomes of xed length [23]. A unique feature of MEP is its ability to store multiple solutions of a problem in a single chromosome. MEP also has an ability to implement code-reuse which is achieved through its representation which allow multiple references to a single sub-structure. This dissertation proposes a new GP algorithm, Improved Gene Expression Programming (IGEP) which im- proves the performance of the traditional GEP by combining the code-reuse capability and simplicity of gene encoding method from MEP and GEP, respectively. The results obtained using the IGEP and the traditional GEP show that the two algorithms are comparable in terms of the success rate when applied on simple problems such as basic logic functions. However, for complex problems such as one-bit Full Adder (FA) and AND-OR Arithmetic Logic Unit (ALU) the IGEP performs better than the traditional GEP due to the code-reuse in IGEP / Mathematical Sciences / M. Sc. (Applied Mathematics)