Computer Engineering Dissertations

Robust design techniques for emerging technologies of computing

Masoud Zamani — Mon, 06 May 2013 11:57:52 PDT

Conventional CMOS technology used in implementation of computational circuits faces major challenges in continues downscaling. Therefore, researchers explore alternative emerging implementation technologies and alternative computational approaches to overcome these challenges. However, in nanoscale regimes due to atomic scale of devices and poor control in nanofabrication, reliability is a major challenge. In this thesis, we will study reliability issues in crossbar nano-architectures, as an example of alternative implementation, as well as reversible logic, as an example of alternative computational technology.

We study two approaches, namely logic mapping and architectural techniques, to incorporate variation and defect tolerance in crossbar nano-architectures. In the...

Bayesian models for unsupervised feature selection

Yue Guan — Wed, 10 Apr 2013 07:10:25 PDT

This dissertation focuses on developing probabilistic models for unsupervised feature selection. High-dimensional data often contain irrelevant and redundant features, which can hurt learning algorithms. One can remove these unwanted features either through removing some subsets of the original features (feature selection) or by transforming data into a lower dimensional feature space. Principal component analysis (PCA) is a popular transformation-based dimensionality reduction method. However, it is not easy to interpret which of the original features are important in PCA. We have designed sparse probabilistic PCA and mixture of sparse probabilistic PCA formulations. By presenting sparse PCA as a probabilistic Bayesian formulation,...

Learning from imperfect and related labels

Yan Yan — Tue, 12 Mar 2013 08:50:25 PDT

Supervised Learning means there is a teacher providing labels or target information given data samples, and the goal is to predict the labels of new or unseen instances. In general, these teachers/labelers may make mistakes. In this thesis, we discuss two supervised learning "imperfect label" scenarios in: (1) multi-label classification and (2) multiple-annotator learning.

Unlike standard classification problems, in multi-label classification, a sample can be assigned to more than one class. Typically, multi-label classifiers are built from multiple binary classifiers one for each class independently. Here, we propose to take advantage of the information that can be derived from the...

Ontology-based reconfigurability of cognitive radio

Leszek Janusz Lechowicz — Tue, 05 Mar 2013 12:40:39 PST

Software Defined Radio (SDR) is a radio in which all or majority of the functionality of the physical layer is implemented in software. One of the biggest benefits of SDR is the relative ease in which the functionality can be changed. Such flexibility could potentially be used in achieving interoperability of the communications systems. Cognitive Radio (CR)—a paradigm that combines SDR technology with a cognitive agent—allows radios to change their operational behaviors in order to achieve a variety of goals, e.g. performance optimization, power saving, opportunistic use of resources, etc. Cognitive Radios are expected to play the leading role in...

Design and implementation of patient specific virtual reality systems for motor rehabilitation

Mark Langdon Sivak — Tue, 05 Mar 2013 12:33:46 PST

There is a need for low-cost, in-home rehabilitation systems with increasing numbers of strokes per year. Therefore a virtual reality system for home use in rehabilitation of the hand and wrist that included hardware, middleware, and software was created and tested with healthy subjects. The Angle Tracking and Location At-home System (ATLAS) was created with potentiome-ter bend sensors to track finger angle and inertial measurement units (IMUs) to track hand orientation and position. To facilitate the communication between the ATLAS and a computer the Device Agnostic Virtual Environment (DAVE) was created as the middleware solution. Patient specific virtual environments have...

Collaborative adaptation of cognitive radio parameters using ontology and policy based approach

Shujun Li — Tue, 16 Oct 2012 09:18:05 PDT

Cognitive radio technology has attracted an increasing interest in academic and industrial communities. One of the motivations of cognitive radio is to enable opportunistic spectrum access through sensing the environment, detecting the underutilized spectrum at a specific time and location, and adjusting the radio's transmission parameters to conform to spectrum utilization regulations and policies. In general, cognitive radio is expected to have the capabilities to (1) sense the environment and collect information of the environment; (2) be aware of the external situation, the internal state and its own capabilities; (3) automatically adapt its parameters and optimize multiple objectives; (4) reason...

p2pSOA: a middleware architecture to enable group collaboration through service composition

Demetris Georgios Galatopoullos — Tue, 22 May 2012 07:19:55 PDT

Digital personal devices, such as Personal Computers and mobile phones, have become pervasive in our times. We more and more depend on their services for our daily tasks from sharing information to managing activities among groups of friends, family, or business peers. With the recent advances in "smart phone" hardware and software technologies it is also now possible to access social networking sites or share content through the Cloud from anywhere at anytime. Moreover, web servers can now be deployed on mobile devices and provide software services to remote users. These new capabilities have far reaching implications since they allow...

Using live virtual machine migration to improve resource efficiency on virtualized data centers

Emmanuel Arzuaga — Tue, 24 Apr 2012 12:08:33 PDT

Efficient resource management is still an open problem in data centers. Although there has been continued improvement in the performance of such large scale systems, the improvements have been mostly at the expense of adding more servers to the system; increasing space requirements and power consumption sometimes without making an effort to better utilize available resources. Modern data centers are growing at a such high pace that space and power consumption are becoming limiting factors. Virtualization has the potential to address these limitations by increasing resource efficiency throughout the data center. Because hardware resources are being shared by different virtual...

Mobile backbones for single and multi-radio networks

Maurizio Antonio Nanni — Tue, 24 Apr 2012 11:48:56 PDT

In this thesis we thoroughly investigate the problem of formulating a distributed protocol for building up and maintaining a backbone structure, i.e., a connected dominating set (CDS) of mobile ad hoc wireless network (MANET). We present a novel solution, termed M-Backs, which shows fast convergence and low overhead. We compared M-Backs with other well-known backbone formation protocols, namely, ETSA [1], TRUNC [2] and GDMAC [3]. Experimental results show that M-Backs outperforms the other solutions in terms of key topological metrics such as backbone size, backbone connectivity, route length and backbone robustness. M-Backs also outperforms the other solutions in delivering an...

K-space undersampling strategies for functional and cardiac MRI: achieving rapid acquisition while maintaining image quality

Onur Afacan — Thu, 15 Mar 2012 12:18:37 PDT

Scan time is the limiting factor in many magnetic resonance imaging applications. One approach to reduce the scan time is undersampling k-space below the Nyquist sampling rate. In this work, for three different MRI applications, we investigate the effects of existing and novel undersampling schemes on both the acquistion time and the resulting image quality. In the first application, a new accelerated multi-shot 3D EPI sequence is proposed to increase the temporal resolution of complex cognitive fMRI studies. A careful combination of two modern acceleration techniques, UNFOLD and GRAPPA, is proposed for use in the secondary phase encoding direction to...

Design methodology based on carbon nanotube field effect transistor (CNFET)

Young Bok Kim — Mon, 09 Jan 2012 12:24:51 PST

This thesis investigates design issues of high speed and low power circuit design using CNTFET Technology. In this thesis modeling and performance benchmarking for nanoscale devices and circuits have been performed for both nanoscale CMOS and carbon nanotube field effect transistor (CNFETs) technologies. Carbon nanotubes with their superior transport properties, excellent thermal conductivities, and high current drivability turned out to be a potential alternative device to the bulk CMOS technology. However, the CNFET technology has new parameters and characteristics which determine the performances such as current driving capability, speed, power consumption and area of circuits. As a result, new design...

Evaluation and enhancement of memory efficiency targeting general-purpose computations on scalable data-parallel GPU architectures

Byunghyun Jang — Mon, 09 Jan 2012 07:37:22 PST

This thesis addresses the memory efficiency of general-purpose applications running on massively multi-threaded, data-parallel GPU architectures. Although scalable, data-parallel GPU architectures and their associated general-purpose programming models offer impressive computational capability and attractive power budgets, the pace of migrating general-purpose applications to this emerging class of architectures is significantly hindered by the efficiency of memory subsystem present on these platforms. Programmers are forced to optimize the memory behavior of their code if they are interested in reaping the full benefits of these high performance, data-parallel architectures.

In this thesis, we present a comprehensive study of memory access behavior for data-parallel...

Precoding and equalization for MIMO broadcast channels with applications in spread spectrum systems

Jin He — Fri, 06 Jan 2012 12:19:00 PST

Proliferation of mobile data applications has increased the demand for wireless communication systems offering high throughput, wide coverage, and improved reliability. The main challenges in the design of such systems are the limited resources---such as constrained transmission power, scarce frequency bandwidth, and limited implementation complexity---and the impairments of the wireless channels, including noise, interference, and fading effects. Multiple-Input Multiple-Output (MIMO) communication has been shown to be one of the most promising emerging wireless technologies that can efficiently boost the data transmission rate, improve system coverage, and enhance link reliability. MIMO is now widely adopted by many mainstream wireless industry standards...

Novel function approximation techniques for large-scale reinforcement learning

Cheng Wu — Tue, 03 May 2011 07:13:47 PDT

Function approximation can be used to improve the performance of reinforcement learners. Traditional techniques, including Tile Coding and Kanerva Coding, can give poor performance when applied to large-scale problems. In our preliminary work, we show that this poor performance is caused by prototype collisions and uneven prototype visit frequency distributions. We describe our adaptive Kanerva-based function approximation algorithm, based on dynamic prototype allocation and adaptation. We show that probabilistic prototype deletion with prototype splitting can make the distribution of visit frequencies more uniform, and that dynamic prototype allocation and adaptation can reduce prototoype collsisions. This approach can significantly improve the...

Adaptive grid computing

Juemin Zhang — Tue, 03 May 2011 05:28:13 PDT

Many innovative scientific applications rely on high-performance computing to perform large computations. Tomosynthesis Mammography, which performs high resolution image reconstruction, is one such computation-intensive application. Currently, it is difficult to launch MPI applications on multiple distributed and heterogeneous computing platforms, or computing grids, since application developers and users must address problems such as application scheduling, resource allocation and co-allocation, and inter-process communication. Our objective is to provide location-, topology-, and administrative-transparent grid computing for MPI applications, while hiding the physical details of computing platforms, and heterogeneous networks from the application developers and users.

In this dissertation, we introduced a resource...

Introducing abstraction to vulnerability analysis

Vilas Keshav Sridharan — Tue, 18 Jan 2011 05:56:12 PST

Tolerance to the effects of transient faults is now a primary design constraint for all major microprocessors. Chip vendors typically set a failure rate target for each design and strive to maximize performance subject to this constraint. To validate that a design meets the failure rate target, vendors perform extensive pre- and post- silicon analysis. One step in this analysis is measuring the Architectural Vulnerability Factor (AVF) of each on-chip structure. The AVF of a hardware structure is the probability that a fault in the structure will affect the output of a program. While AVF generates meaningful insight into system...

Improving large margin classifiers using relationships among samples

Volkan Vural — Thu, 09 Dec 2010 10:30:51 PST

Support vector machine (SVM) is a powerful supervised classification algorithm that has been successful in many real-world problems such as text categorization, face recognition, and applications in bioinformatics and computer-aided diagnosis. Although SVM is popular and accurate, it has some limitations as well. In this thesis, we focus on three major limitations of SVM and introduce various algorithms that utilize the relationships among samples to overcome these issues. Firstly, a limitation of SVM and that of supervised learning algorithms in general is that they only learn from labeled data. However, in many domains, labeled insances are typically costly to obtain....

DNA self-assembly for efficient error-tolerant nanomanufacturing

Masoud Hashempour — Tue, 27 Apr 2010 10:23:04 PDT

Biology makes things far smaller and more complex than anything produced by human engineering. The biotechnology revolution has for the first time given us the tools necessary to consider engineering on the molecular level. Research in DNA computation, launched by Len Adleman, has opened the door for experimental study of programmable biochemical reactions. In this thesis we focus on a single biochemical mechanism, the self-assembly of DNA structures, that is theoretically sufficient for Turing-universal computation. Wang tiles theory shows how jigsaw-shaped tiles can simulate the operation of a Turing machine. The theory combines Wangs purely mathematical Tiling Problem with the...

System-level memory power and performance optimization for system-on-a-chip embedded systems

Ke Ning — Fri, 18 Dec 2009 11:53:27 PST

Power has become a first rate design issue in microprocessor design. Power efficiency is especially critical for battery-powered embedded systems. Technology trends are making data communication, both on-chip and off-chip, more expensive relative to computation. Evaluating power-performance design trade-offs at the architectural level still requires more research study. In this dissertation, we will show how microprocessor power, especially in the memory sub-system, is consumed during program execution. We also show that the external memory system in a low power System-on-a-Chip (SOC) embedded system has significant impact on overall system power. The source of memory power consumption is due to the...

Investigating the utility of software semantics for host-based intrusion detection systems

Micha Moffie — Fri, 18 Dec 2009 11:53:26 PST

Zero day attacks and hidden Malware pose a grave threat to computer users. To date, widespread security measures such as anti-virus packages and firewalls have proven to be ineffective in guarding against these types of malware. New security measures are essential to secure computer systems, protect digital information, and restore user confidence. Security mechanisms which are able to differentiate regular (normal) behavior from malicious (abnormal) behavior promise new ways to effectively detect, counter and ultimately prevent the execution of zero day attacks and hidden malware. In this thesis we explore the utility of different software semantics for detecting malicious behavior....