Thesis (Ph.D.)--York University, 2008. Graduate Programme in Communication and Culture. / Typescript. Includes bibliographical references (leaves 269-283). Also available on the Internet. MODE OF ACCESS via web browser by entering the following URL: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&res_dat=xri:pqdiss&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&rft_dat=xri:pqdiss:NR46001

This dissertation is on sonic surveillance in the neoliberal context and its implication for privacy, agency, sovereignty, ownership and control. This research focuses on the social, political and ethical conceptions of privacy through musical consumption and sonic practices in the United States. I investigate non-creative recording practices in neoliberal life and identify the listening practices of surveillance capitalism to better understand how power circulates through sound. Through a multi-sited ethnography, I conduct three case studies on the recording and listening capacity of technological devices of everyday life in order to theorize what I term “the neoliberal ear”– a twenty-first century mode of listening to the world embedded into surveillance capitalism. I analyze three sonic tools of surveillance capitalism: streaming service Spotify, Smart Home device Amazon Echo, and Smart City communication hub LinkNYC. These technologies, I argue, embody and promote neoliberal ideology, and the companies that produces them operate within a neoliberal mode of governance allowed by public policies. This dissertation is interdisciplinary in scope and operates at theoretical crossings of sound and power, technology and cultural practices, and disciplinary crossings of music, law and computer science. I draw from, and build connections between; sound studies, ethnomusicology, legal literature and scholarship on copyright and privacy, surveillance studies, science and technology studies, and discourses on AI and ethics, to form theories of sound and power in surveillance capitalism.

Thesis (M.S.)--Rochester Institute of Technology, 2008. / Typescript. Includes bibliographical references (leaves 22-25).

This thesis consists of three papers in optimization and revenue management over complex networks: Robust Linear Control in Transmission Systems, Online Learning and Optimization Under a New Linear-Threshold Model with Negative Influence, and Revenue Management with Complementarity Products. This thesis contributes to analytical methods for optimization problems in complex networks, namely, power network, social network and product network. In Chapter 2, we describe a robust multiperiod transmission planning model including renewables and batteries, where battery output is used to partly offset renewable output deviations from forecast. A central element is a nonconvex battery operation model plus a robust model of forecast errors and a linear control scheme. Even though the problem is nonconvex we provide an efficient and theoretically valid algorithm that effectively solves cases on large transmission systems. In Chapter 3, we propose a new class of Linear Threshold Model-based information-diffusion model that incorporates the formation and spread of negative attitude. We call such models negativity-aware. We show that in these models, the expected positive influence is a monotone sub-modular function of the seed set. Thus we can use a greedy algorithm to construct a solution with constant approximation guarantee when the objective is to select a seed set of fixed size to maximize positive influence. Our models are flexible enough to account for both the features of local users and the features of the information being propagated in the diffusion. We analyze an online-learning setting for a multi-round influence-maximization problem, where an agent is actively learning the diffusion parameters over time while trying to maximize total cumulative positive influence. We develop a class of online learning algorithms and provide the theoretical upper bound on the regret. In Chapter 4, we propose a tractable information-diffusion-based framework to capture complementary relationships among products. Using this framework, we investigate how various revenue-management decisions can be optimized. In particular, we prove that several fundamental problems involving complementary products, such as promotional pricing, product recommendation, and category planning, can be formulated as sub-modular maximization problems, and can be solved by tractable greedy algorithms with guarantees on the quality of the solutions. We validate our model using a dataset that contains product reviews and metadata from Amazon from May 1996 to July 2014. We also analyze an online-learning setting for revenue-maximization with complementary products. In this setting, we assume that the retailer has access only to sales observations. That is, she can only observe whether a product is purchased from her. This assumption leads to diffusion models with novel node-level feedback, in contrast to classical models that have edge-level feedback. We conduct confidence region analysis on the maximum likelihood estimator for our models, develop online-learning algorithms, and analyze their performance in both theoretical and practical perspectives.

The past decade has seen tremendous growth in potential applications of language technologies in our daily lives due to increasing data, computational resources, and user interfaces. An important step to support emerging applications is the development of algorithms for processing the rich variety of human-generated text and extracting relevant information. Machine learning, especially deep learning, has seen increasing success on various text benchmarks. However, while standard benchmarks have static tasks with expensive human-labeled data, real-world applications are characterized by dynamic task specifications and limited resources for data labeling, thus making it challenging to transfer the success of supervised machine learning to the real world. To deploy language technologies at scale, it is crucial to develop alternative techniques for teaching machines beyond data labeling. In this dissertation, we address this data labeling bottleneck by studying and presenting resource-efficient frameworks for teaching machine learning models to solve language tasks across diverse domains and languages. Our goal is to (i) support emerging real-world problems without the expensive requirement of large-scale manual data labeling; and (ii) assist humans in teaching machines via more flexible types of interaction. Towards this goal, we describe our collaborations with experts across domains (including public health, earth sciences, news, and e-commerce) to integrate weakly-supervised neural networks into operational systems, and we present efficient machine teaching frameworks that leverage flexible forms of declarative knowledge as supervision: coarse labels, large hierarchical taxonomies, seed words, bilingual word translations, and general labeling rules. First, we present two neural network architectures that we designed to leverage weak supervision in the form of coarse labels and hierarchical taxonomies, respectively, and highlight their successful integration into operational systems. Our Hierarchical Sigmoid Attention Network (HSAN) learns to highlight important sentences of potentially long documents without sentence-level supervision by, instead, using coarse-grained supervision at the document level. HSAN improves over previous weakly supervised learning approaches across sentiment classification benchmarks and has been deployed to help inspections in health departments for the discovery of foodborne illness outbreaks. We also present TXtract, a neural network that extracts attributes for e-commerce products from thousands of diverse categories without using manually labeled data for each category, by instead considering category relationships in a hierarchical taxonomy. TXtract is a core component of Amazon’s AutoKnow, a system that collects knowledge facts for over 10K product categories, and serves such information to Amazon search and product detail pages. Second, we present architecture-agnostic machine teaching frameworks that we applied across domains, languages, and tasks. Our weakly-supervised co-training framework can train any type of text classifier using just a small number of class-indicative seed words and unlabeled data. In contrast to previous work that use seed words to initialize embedding layers, our iterative seed word distillation (ISWD) method leverages the predictive power of seed words as supervision signals and shows strong performance improvements for aspect detection in reviews across domains and languages. We further demonstrate the cross-lingual transfer abilities of our co-training approach via cross-lingual teacher-student (CLTS), a method for training document classifiers across diverse languages using labeled documents only in English and a limited budget for bilingual translations. Not all classification tasks, however, can be effectively addressed using human supervision in the form of seed words. To capture a broader variety of tasks, we present weakly-supervised self-training (ASTRA), a weakly-supervised learning framework for training a classifier using more general labeling rules in addition to labeled and unlabeled data. As a complete set of accurate rules may be hard to obtain all in one shot, we further present an interactive framework that assists human annotators by automatically suggesting candidate labeling rules. In conclusion, this thesis demonstrates the benefits of teaching machines with different types of interaction than the standard data labeling paradigm and shows promising results for new applications across domains and languages. To facilitate future research, we publish our code implementations and design new challenging benchmarks with various types of supervision. We believe that our proposed frameworks and experimental findings will influence research and will enable new applications of language technologies without the costly requirement of large manually labeled datasets.

Networked systems integrating software with the physical world are known as cyber-physical systems (CPSs). CPSs have been used in diverse sectors, including power generation and distribution, transportation, industrial systems, and building management. The diversity of applications and interdisciplinary nature make CPSs exciting to design and build but challenging to manage once deployed. Deployed CPSs must adapt to changes in the operating environment or the system's architecture, e.g., when outdated or malfunctioning components need to be replaced. Skilled human operators have traditionally performed such adaptations using centralized management protocols. As the CPS grows, management tasks become more complex, tedious, and error-prone. This dissertation studies management challenges in deployed CPSs. It is based on practical research with CPSs of various sizes and diverse application domains, from the large geographically dispersed electrical grid to small-scale consumer Internet of Things (IoT) systems. We study the management challenges unique to each system and propose network services and protocols specifically designed to reduce the amount of management overhead, drawing inspiration from autonomic systems and networking research. We first introduce PhoenixSEN, a self-managing ad hoc network designed to restore connectivity in the electrical grid after a large-scale outage. The electrical grid is a large, heterogeneous, geographically dispersed CPS. We analyze the U.S. electrical grid network subsystem, propose an ad hoc network to temporarily replace the network subsystem during a blackout, and discuss the experimental evaluation of the network on a one-of-a-kind physical electrical grid testbed. The novel aspects of PhoenixSEN lie in a combination of existing and new network technologies and manageability by power distribution industry operators. Motivated by the challenges of running unmodified third-party applications in an ad hoc network like PhoenixSEN, we propose a geographic resource discovery and query processing service for federated CPSs called SenSQL. The service combines a resource discovery protocol inspired by the LoST protocol with a standard SQL-based query interface. SenSQL aims to simplify the development of applications for federated or administratively decoupled autonomous cyber-physical systems without a single administrative or technological point of failure. The SenSQL framework balances control over autonomous cyber-physical devices and their data with service federation, limiting the application's reliance on centralized infrastructures or services. We conclude the first part of the dissertation by presenting the design and implementation of a testbed for usability experiments with mission-critical voice, a vital communication modality in PhoenixSEN, and during emergency scenarios in general. The testbed can be used to conduct human-subject studies under emulated network conditions to assess the influence of various network parameters on the end-user's quality of experience. The second dissertation part focuses on network enrollment of IoT devices, a management process that is often complicated, frustrating, and error-prone, particularly in consumer-oriented systems. We motivate the work by reverse-engineering and analyzing Amazon Echo's network enrollment protocol. The Echo is one of the most widely deployed IoT devices and, thus, an excellent case study. We learn that the process is rather complicated and cumbersome. We then present a systematic study of IoT network enrollment with a focus on consumer IoT devices in advanced deployment scenarios, e.g., third-party installations, shared physical spaces, or evolving IoT systems. We evaluate existing frameworks and their shortcoming and propose WIDE, a network-independent enrollment framework designed to minimize user interactions to enable advanced deployment scenarios. WIDE is designed for large-scale or heterogeneous IoT systems where multiple independent entities cooperate to set the system up. We also discuss the design of a human-subject study to compare and contrast the usability of network enrollment frameworks. A secure network must authenticate a new device before it can be enrolled. The authentication step usually requires physical device access, which may be impossible in many advanced deployment scenarios, e.g., when IoT devices are installed by a specialist in physically unreachable locations. We propose Lighthouse, a visible-light authentication protocol for physically inaccessible IoT devices. We discuss the protocol's design, develop transmitter and receiver prototypes, and evaluate the system. Our measurements with off-the-shelf components over realistic distances indicate authentication times shorter or comparable with existing methods involving gaining physical access to the device. We also illustrate how the visible-light authentication protocol could be used as another authentication method in other network enrollment frameworks.