Theses and Dissertations from UMD
Permanent URI for this communityhttp://hdl.handle.net/1903/2
New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
4 results
Search Results
Item Mutual Information-based RBM Neural Networks(2016) Peng, Kang-Hao; Chellappa, Rama; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)(Deep) neural networks are increasingly being used for various computer vision and pattern recognition tasks due to their strong ability to learn highly discriminative features. However, quantitative analysis of their classication ability and design philosophies are still nebulous. In this work, we use information theory to analyze the concatenated restricted Boltzmann machines (RBMs) and propose a mutual information-based RBM neural networks (MI-RBM). We develop a novel pretraining algorithm to maximize the mutual information between RBMs. Extensive experimental results on various classication tasks show the eectiveness of the proposed approach.Item Parallel Computation of Nonrigid Image Registration(2011) Leung, Frances Kimpik; Shekhar, Raj; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Automatic intensity-based nonrigid image registration brings significant impact in medical applications such as multimodality fusion of images, serial comparison for monitoring disease progression or regression, and minimally invasive image-guided interventions. However, due to memory and compute intensive nature of the operations, intensity-based image registration has remained too slow to be practical for clinical adoption, with its use limited primarily to as a pre-operative too. Efficient registration methods can lead to new possibilities for development of improved and interactive intraoperative tools and capabilities. In this thesis, we propose an efficient parallel implementation for intensity-based three-dimensional nonrigid image registration on a commodity graphics processing unit. Optimization techniques are developed to accelerate the compute-intensive mutual information computation. The study is performed on the hierarchical volume subdivision-based algorithm, which is inherently faster than other nonrigid registration algorithms and structurally well-suited for data-parallel computation platforms. The proposed implementation achieves more than 50-fold runtime improvement over a standard implementation on a CPU. The execution time of nonrigid image registration is reduced from hours to minutes while retaining the same level of registration accuracy.Item Temporal dynamics of MEG phase information during speech perception: Segmentation and neural communication using mutual information and phase locking(2011) Cogan, Gregory Brendan; Idsardi, William; Neuroscience and Cognitive Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The incoming speech stream contains a rich amount of temporal information. In particular, information on slow time scales, the delta and theta band (125 - 1000 ms, 1 - 8 Hz), corresponds to prosodic and syllabic information while information on faster time scales (20-40 ms, 25 - 50 Hz) corresponds to feature/phonemic information. In order for speech perception to occur, this signal must be segregated into meaningful units of analysis and then processed in a distributed network of brain regions. Recent evidence suggests that low frequency phase information in the delta and theta bands of the Magnetoencephalography (MEG) signal plays an important role for tracking and segmenting the incoming signal into units of analysis. This thesis utilized a novel method of analysis, Mutual Information (MI) to characterize the relative information contributions of these low frequency phases. Reliable information pertaining to the stimulus was present in both delta and theta bands (3 - 5 Hz, 5 - 7 Hz) and information within each of these three sub-bands was independent of each other. A second experiment demonstrated that the information present in these bands differed significantly for speech and a non-speech control condition, suggesting that contrary to previous results, a purely acoustic hypothesis of this segmentation is not supported. A third experiment found that both low (delta and theta) and high (gamma) frequency information is utilized to facilitate communication between brain areas thought to underlie speech perception. Distinct auditory/speech networks that operated exclusively using these frequencies were revealed, suggesting a privileged role for these timescales for neural communication between brain regions. Taken together these results suggest that timescales that correspond linguistically to important aspects of the speech stream also facilitate segmentation of the incoming signal and communication between brain areas that perform neural computation.Item Parallelization of Non-Rigid Image Registration(2008) Philip, Mathew; Shekhar, Raj; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Non-rigid image registration finds use in a wide range of medical applications ranging from diagnostics to minimally invasive image-guided interventions. Automatic non-rigid image registration algorithms are computationally intensive in that they can take hours to register two images. Although hierarchical volume subdivision-based algorithms are inherently faster than other non-rigid registration algorithms, they can still take a long time to register two images. We show a parallel implementation of one such previously reported and well tested algorithm on a cluster of thirty two processors which reduces the registration time from hours to a few minutes. Mutual information (MI) is one of the most commonly used image similarity measures used in medical image registration and also in the mentioned algorithm. In addition to parallel implementation, we propose a new concept based on bit-slicing to accelerate computation of MI on the cluster and, more generally, on any parallel computing platform such as the Graphics processor units (GPUs). GPUs are becoming increasingly common for general purpose computing in the area of medical imaging as they can execute algorithms faster by leveraging the parallel processing power they offer. However, the standard implementation of MI does not map well to the GPU architecture, leading earlier investigators to compute only an inexact version of MI on the GPU to achieve speedup. The bit-slicing technique we have proposed enables us to demonstrate an exact implementation of MI on the GPU without adversely affecting the speedup.