published_or_final_version / Computer Science / Master / Master of Philosophy

Thesis (M. Phil.)--University of Hong Kong, 1994. / Includes bibliographical references (leaves 100-101).

Thesis (Ph.D)--Computing, Georgia Institute of Technology, 2010. / Committee Chair: Ramachandran, Umakishore; Committee Member: Clark, Nathan; Committee Member: Haskin, Roger; Committee Member: Pu, Calton; Committee Member: Rehg, James. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Specifying varying speeds and temporal relationships is necessary when programming graphical animations, but support for temporal programming has usually been done by adding new language features to a Visual Programming Language (VPL), and these features must be mastered over and above the other aspects of the VPL. However, some researchers have believed that time should be able to be treated like just another dimension. In this thesis, we explore whether temporal programming can indeed be done using exactly the same devices as in spatial programming in grid-oriented VPLs. Toward this end, we provide a continuum of models aimed at this goal and discuss their advantages and disadvantages. Also, we identify core issues that help illuminate the essence of the problem. / Graduation date: 2001

Thesis (Ph.D)--Computing, Georgia Institute of Technology, 2010. / Committee Chair: Dellaert, Frank; Committee Co-Chair: Rehg, James; Committee Member: Bobick, Aaron; Committee Member: Essa, Irfan; Committee Member: Smyth, Padhraic. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Continuous live stream analysis applications are increasingly common. Video-based surveillance, emergency response, disaster recovery, and critical infrastructure monitoring are all examples of such applications. These applications are distributed and typically require significant computing resources (like a cluster of workstations) for analysis. In addition to live data, many such applications also require access to historical data that was streamed in the past and is now archived. While distributed programming support for traditional high-performance computing applications is fairly mature, existing solutions for live stream analysis applications are still in their early stages and, in our view, inadequate. We explore the system-level value of recognizing temporal properties -- a critical aspect of the application domain. We present "temporal streams", a programming model supporting a higher-level, domain-targeted programming abstraction for such applications. It provides a simple but expressive stream abstraction encompassing transport, manipulation and storage of streaming data. The semantics of the programming model are tailored to the application domain by explicitly recognizing the temporal aspects of continuous streams, providing a common interface for both time-based retrieval of current streaming data and data persistence. The unifying trait of time enables access to both current streaming data and archived historical data using the same interface; the communication and storage abstraction are the same -- a unified stream data abstraction, uniformly modeling stream data interactions. "Temporal streams" defines how distributed threads of computation interact implicitly via streams, but does not impose a particular model of computation constraining the interactions between distributed actors, targeting loosely coupled distributed systems with no centralized control. In particular, it targets stream analysis scenarios requiring significant signal processing on heavyweight streams such as audio and video. These unstructured streams are data rich but are not directly interpretable until meaningful features are extracted; consequently, feature detection and subsequent analysis are the major computational requirements. We also use the programming model as a vehicle for exploring systems software design issues, realizing "temporal streams" as a distributed runtime in the tradition of loosely coupled distributed systems with strong communication boundaries. We thoroughly examine the concrete software architecture and elements of implementation. We also describe two generations of system implementations, including the broad development philosophy, specific design principles and salient low-level details. The runtime is designed to be relatively lightweight and suitable as a substrate for higher-level, more domain-specific middleware or application functionality. Even with a relatively simple programming model, a carefully designed system architecture can provide a surprisingly rich and flexibly substrate for upper software layers. We also evaluate our system implementation in two ways; first, we present a series of quantitative experimental results designed to assess the performance of key primitives in our architecture in isolation. We also use motivating applications to evaluate "temporal streams" in the context of realistic application scenarios. We develop three motivating applications and provide quantitative and qualitative analyses of these applications in the context of "temporal streams." We show that, although it provides needed higher-level functionality to enable live stream analysis applications, our runtime does not add significant overhead to the stream computation at the core of each application. Finally, we also review the relationship of "temporal streams" (both the programming model and architecture) to other approaches, including database-oriented Stream Data Management Systems (SDMS), various stream processing engines, stream programming languages and parallel batch processing systems, as well as traditional distributed programming systems and communication frameworks.

Automated analysis of temporal data is a task of utmost importance for intelligent machines. For example, ubiquitous computing systems need to understand the intention of humans from the stream of sensory information, and health-care monitoring systems can assist patients and doctors by providing automatically annotated daily health reports. We present a set of extensions of switching linear dynamic systems (SLDSs) which provide the ability to capture the higher-order temporal structures within data and to produce more accurate results for the tasks such as labeling and estimation of global variations within data. The presented models are formulated within a dynamic Bayesian network formulation along with the inference and learning methods thereof. First, segmental SLDSs (S-SLDSs) produce superior labeling results by capturing the descriptive duration patterns within each LDS segment. The encoded duration models describe data more descriptively and allow us to avoid the severe problem of over-segmented labels, which leads to superior accuracy. Second, parametric SLDSs (P-SLDSs) allows us to encode the temporal data with global variations. In particular, we have identified two types of global systematic variations : temporal and spatial variations. The P-SLDS model assumes that there is an underlying canonical model which is globally transformed in time and space by the two associated global parameters respectively. Third, we present hierarchical SLDSs (H-SLDSs), a generalization of standard SLDSs with hierarchic Markov chains. H-SLDSs are able to encode temporal data which exhibits hierarchic structure where the underlying low-level temporal patterns repeatedly appear among different higher-level contexts. The developed SLDS extensions have been applied to two real-world problems. The first problem is to automatically decode the dance messages of honey bee dances where the goal is to correctly segment the dance sequences into different regimes and parse the messages about the location of food sources embedded in the data. The second problem is to analyze wearable exercise data where we aim to provide an automatically generated exercise record at multiple temporal and semantic resolutions. It is demonstrated that the H-SLDS model with multiple layers can be learned from data, and can be successfully applied to interpret the exercise data at multiple granularities.