Computer Science
http://hdl.handle.net/10012/9930
2017-04-26T08:01:59ZMulti-Agent Modeling of Risk-Aware and Privacy-Preserving Recommender Systems
http://hdl.handle.net/10012/11732
Multi-Agent Modeling of Risk-Aware and Privacy-Preserving Recommender Systems
Srivastava, Vishnu
Recent progress in the field of recommender systems has led to increases in the accuracy and significant improvements in the personalization of recommendations. These results are being achieved in general by gathering more user data and generating relevant insights from it. However, user privacy concerns are often underestimated and recommendation risks are not usually addressed. In fact, many users are not sufficiently aware of what data is collected about them and how the data is collected (e.g., whether third parties are collecting and selling their personal information).
Research in the area of recommender systems should strive towards not only achieving high accuracy of the generated recommendations but also protecting the user’s privacy and making recommender systems aware of the user’s context, which involves the user’s intentions and the user’s current situation. Through research it has been established that a tradeoff is required between the accuracy, the privacy and the risks in a recommender system and that it is highly unlikely to have recommender systems completely satisfying all the context-aware and privacy-preserving requirements. Nonetheless, a significant attempt can be made to describe a novel modeling approach that supports designing a recommender system encompassing some of these previously mentioned requirements.
This thesis focuses on a multi-agent based system model of recommender systems by introducing both privacy and risk-related abstractions into traditional recommender systems and breaking down the system into three different subsystems. Such a description of the system will be able to represent a subset of recommender systems which can be classified as both risk-aware and privacy-preserving. The applicability of the approach is illustrated by a case study involving a job recommender system in which the general design model is instantiated to represent the required domain-specific abstractions.
2017-04-25T00:00:00ZSoftware Engineering for Big Data Systems
http://hdl.handle.net/10012/11721
Software Engineering for Big Data Systems
Kumar, Vijay Dipti
Software engineering is the application of a systematic approach to designing, operating and maintaining software systems and the study of all the activities involved in achieving the same. The software engineering discipline and research into software systems flourished with the advent of computers and the technological revolution ushered in by the World Wide Web and the Internet. Software systems have grown dramatically to the point of becoming ubiquitous. They have a significant impact on the global economy and on how we interact and communicate with each other and with computers using software in our daily lives.
However, there have been major changes in the type of software systems developed over the years. In the past decade owing to breakthrough advancements in cloud and mobile computing technologies, unprecedented volumes of hitherto inaccessible data, referred to as big data, has become available to technology companies and business organizations farsighted and discerning enough to use it to create new products, and services generating astounding profits. The advent of big data and software systems utilizing big data has presented a new sphere of growth for the software engineering discipline. Researchers, entrepreneurs and major corporations are all looking into big data systems to extract the maximum value from data available to them. Software engineering for big data systems is an emergent field that is starting to witness a lot of important research activity.
This thesis investigates the application of software engineering knowledge areas and standard practices, established over the years by the software engineering research community, into developing big data systems by:
- surveying the existing software engineering literature on applying software engineering principles into developing and supporting big data systems;
- identifying the fields of application for big data systems;
- investigating the software engineering knowledge areas that have seen research related
to big data systems;
- revealing the gaps in the knowledge areas that require more focus for big data systems
development; and
- determining the open research challenges in each software engineering knowledge area
that need to be met.
The analysis and results obtained from this thesis reveal that recent advances made in
distributed computing, non-relational databases, and machine learning applications have
lured the software engineering research and business communities primarily into focusing
on system design and architecture of big data systems. Despite the instrumental role
played by big data systems in the success of several businesses organizations and technology
companies by transforming them into market leaders, developing and maintaining stable,
robust, and scalable big data systems is still a distant milestone. This can be attributed
to the paucity of much deserved research attention into more fundamental and equally important
software engineering activities like requirements engineering, testing, and creating
good quality assurance practices for big data systems.
2017-04-24T00:00:00ZProperties of Two-Dimensional Words
http://hdl.handle.net/10012/11714
Properties of Two-Dimensional Words
Smith, Taylor
Combinatorics on words in one dimension is a well-studied subfield of theoretical computer science with its origins in the early 20th century. However, the closely-related study of two-dimensional words is not as popular, even though many results seem naturally extendable from the one-dimensional case. This thesis investigates various properties of these two-dimensional words.
In the early 1960s, Roger Lyndon and Marcel-Paul Schutzenberger developed two famous results on conditions where nontrivial prefixes and suffixes of a one-dimensional word are identical and on conditions where two one-dimensional words commute. Here, the theorems of Lyndon and Schutzenberger are extended in the one-dimensional case to include a number of additional equivalent conditions. One such condition is shown to be equivalent to the defect theorem from formal languages and coding theory. The same theorems of Lyndon and Schutzenberger are then generalized to the two-dimensional case.
The study of two-dimensional words continues by considering primitivity and periodicity in two dimensions, where a method is developed to enumerate two-dimensional primitive words. An efficient computer algorithm is presented to assist with checking the property of primitivity in two dimensions. Finally, borders in both one and two dimensions are considered, with some results being proved and others being offered as suggestions for future work. Another efficient algorithm is presented to assist with checking whether a two-dimensional word is bordered.
The thesis concludes with a selection of open problems and an appendix containing extensive data related to one such open problem.
2017-04-21T00:00:00ZExtended Nonlocal Games
http://hdl.handle.net/10012/11620
Extended Nonlocal Games
Russo, Vincent
The notions of entanglement and nonlocality are among the most striking ingredients found in quantum information theory. One tool to better understand these notions is the model of nonlocal games; a mathematical framework that abstractly models a physical system. The simplest instance of a nonlocal game involves two players, Alice and Bob, who are not allowed to communicate with each other once the game has started and who play cooperatively against an adversary referred to as the referee.
The focus of this thesis is a class of games called extended nonlocal games, of which nonlocal games are a subset. In an extended nonlocal game, the players initially share a tripartite state with the referee. In such games, the winning conditions for Alice and Bob may depend on outcomes of measurements made by the referee, on its part of the shared quantum state, in addition to Alice and Bob's answers to the questions sent by the referee.
We build up the framework for extended nonlocal games and study their properties and how they relate to nonlocal games. In doing so, we study the types of strategies that Alice and Bob may adopt in such a game. For instance, we refer to strategies where Alice and Bob use quantum resources as standard quantum strategies and strategies where there is an absence of entanglement as an unentangled strategy. These formulations of strategies are purposefully reminiscent of the respective quantum and classical strategies that Alice and Bob use in a nonlocal game, and we also consider other types of strategies with a similar correspondence for the class of extended nonlocal games.
We consider the value of an extended nonlocal game when Alice and Bob apply a particular strategy, again in a similar manner to the class of nonlocal games. Unlike computing the unentangled value where tractable algorithms exist, directly computing the standard quantum value of an extended nonlocal game is an intractable problem. We introduce a technique that allows one to place upper bounds on the standard quantum value of an extended nonlocal game. Our technique is a generalization of what we refer to as the QC hierarchy which was studied independently in works by Doherty, Liang, Toner, and Wehner as well as by Navascues, Pironio, and Acin. This technique yields an upper bound approximation for the quantum value of a nonlocal game.
We also consider the question of whether or not the dimensionality of the state that Alice and Bob share as part of their standard quantum strategy makes any difference in how well they can play the game. That is, does there exist an extended nonlocal game where Alice and Bob can win with a higher probability if they share a state where the dimension is infinite? We answer this question in the affirmative and provide a specific example of an extended nonlocal game that exhibits this behavior.
We study a type of extended nonlocal game referred to as a monogamy-of-entanglement game, introduced by Tomamichel, Fehr, Kaniewski, and Wehner, and present a number of new results for this class of game. Specifically, we consider how the standard quantum value and unentangled value of these games relate to each other. We find that for certain classes of monogamy-of-entanglement games, Alice and Bob stand to gain no benefit in using a standard quantum strategy over an unentangled strategy, that is, they perform just as well without making use of entanglement in their strategy. However, we show that there does exist a monogamy-of-entanglement game in which Alice and Bob do perform strictly better if they make use of a standard quantum strategy. We also analyze the parallel repetition of monogamy-of-entanglement games; the study of how a game performs when there are multiple instances of the game played independently. We find that certain classes of monogamy-of-entanglement games obey strong parallel repetition. In contrast, when Alice and Bob use a non-signaling strategy in a monogamy-of-entanglement game, we find that strong parallel repetition is not obeyed.
2017-03-31T00:00:00Z