This work addresses neural and analog computation on reconfigurable mixed-signal platforms. Many engineered systems could gain tremendous benefits by emulating neural systems. For example, neural systems are incredibly power efficient and fault-tolerant. They are also capable of types of computation that we cannot yet match with conventional computers. Neuromorphic engineers typically implement neural computation using analog circuits because they are low-power and naturally model some aspects of neurobiology. One problem with analog circuits is that they are typically inflexible. To address this shortcoming, our lab has developed reconfigurable analog systems known as Field Programmable Analog Arrays (FPAAs). This dissertation consists of two main parts. The first is the implementation of neural and analog circuits on FPAAs. We first implemented an adaptive winner-take-all circuit, which could model attention in neural systems. Next, we modeled the dendrite, which is the conductive tissue that relays inputs from synapses to the neuron cell body. We also implemented a subtractive music synthesizer, perhaps providing the electronic music synthesis community with a good platform for experimentation. Finally, we conducted a number of neural learning experiments on a neuromorphic platform. The second part of this dissertation includes design aspects of new FPAAs, including configurable blocks that can be used as current-mode DACs in a digitally-enhanced FPAA, the RASP 2.9v. We also consider the design of a new neuromorphic platform containing 256 neurons and over 200,000 synapses, many with learning capability. We also created an active delay line that could be used for beamforming or FIR filter applications. In summary, this work adds to the field of reconfigurable systems by both showing how to implement circuits with them and creating new systems based on lessons learned while working with previous systems.

Cancer, the uncontrolled growth of morphologically and genetically abnormal cells in the body, is a major worldwide public health problem and there is a great need for novel insights into this disease. The majority of tumors arise from the acquisition of somatic alterations leading to changes in gene function and expression. The clinical success of targeted therapeutics in molecularly defined subsets of patients has highlighted the need for comprehensive characterization of the somatic alterations in individual cancer types. Copy number profiling using SNP arrays is a common approach for profiling the extent of copy number variation across the cancer genome. In addition, next-generation sequencing technologies now offer researchers the ability to also systematically catalog nucleotide substitutions and structural rearrangements in dramatically less time and expense. In this thesis, we describe the application of SNP arrays and whole-exome sequencing to characterize two separate cohorts of cancer samples, as well as describe the development of a software tool to aid in the annotation of mutational data. Specifically, we detailed focal amplifications of PDGFRA and KIT in a combined set of lung adenocarcinoma and squamous cell carcinomas. Furthermore, in a cohort of small bowel neuroendocrine tumors, we characterized the global genetic landscape to show that these tumors are molecularly distinct from other neuroendocrine tumors. Lastly, we report Oncotator, a novel web application and service for comprehensive annotation of point mutations and indels found in cancer. It is hoped that the knowledge gained from these studies will fuel improvements in cancer diagnosis, prognosis, and therapy.

Not available / text

Not available / text

Not available / text

Not available / text

published_or_final_version / Industrial and Manufacturing Systems Engineering / Master / Master of Philosophy

Plenoptic cameras and Shack-Hartmann wavefront sensors are lenslet-based optical systems that do not form a conventional image. The addition of a lens array into these systems allows for the aberrations generated by the combination of the object and the optical components located prior to the lens array to be measured or corrected with post-processing. This dissertation provides a ray selection method to determine the rays that pass through each lenslet in a lenslet-based system. This first-order, ray trace method is developed for any lenslet-based system with a well-defined fore optic, where in this dissertation the fore optic is all of the optical components located prior to the lens array. For example, in a plenoptic camera the fore optic is a standard camera lens. Because a lens array at any location after the exit pupil of the fore optic is considered in this analysis, it is applicable to both plenoptic cameras and Shack-Hartmann wavefront sensors. Only a generic, unaberrated fore optic is considered, but this dissertation establishes a framework for considering the effect of an aberrated fore optic in lenslet-based systems. The rays from the fore optic that pass through a lenslet placed at any location after the fore optic are determined. This collection of rays is reduced to three rays that describe the entire lenslet ray set. The lenslet ray set is determined at the object, image, and pupil planes of the fore optic. The consideration of the apertures that define the lenslet ray set for an on-axis lenslet leads to three classes of lenslet-based systems. Vignetting of the lenslet rays is considered for off-axis lenslets. Finally, the lenslet ray set is normalized into terms similar to the field and aperture vector used to describe the aberrated wavefront of the fore optic. The analysis in this dissertation is complementary to other first-order models that have been developed for a specific plenoptic camera layout or Shack-Hartmann wavefront sensor application. This general analysis determines the location where the rays of each lenslet pass through the fore optic establishing a framework to consider the effect of an aberrated fore optic in a future analysis.

No description available.

During the past two decades, techniques and devices were developed to transmit and receive signals with a phased array instead of a single coil in the MRI (Magnetic Resonance Imaging) system. The two techniques to simultaneously transmit and receive RF signals using phased arrays are called parallel excitation (pTx) and parallel imaging (PI), respectively. These two techniques lead to shorter transmit pulses for higher imaging quality and faster data acquisition correspondingly. This dissertation focuses on improving the efficiency of the pTx pulse design and the PI reconstruction in MRI. Both PI and pTx benefit from the increased number of elements of the array. However, efficiency concerns may arise which include: (1) In PI, the computation cost of the reconstructions and the achievable acceleration factors and (2) in pTx, the pulse design speed and memory cost. The work presented in this dissertation addresses these issues. First, a correlation based channel reduction algorithm is developed to reduce the computation cost of PI reconstruction. In conventional k-domain methods, the individual channel data is reconstructed via linear interpolation of the neighbourhood data from all channels. In this proposed algorithm, we choose only a subset of the channels based on the spatial correlation. The results have shown that the computation cost can be significantly reduced with similar or higher reconstruction accuracy. Then, a new parallel imaging method named parallel imaging using localized receive arrays with Sinc interpolation(PILARS) is proposed to improve the actual acceleration factor and to reduce the computation cost. It employs the local support of individual coils and pre-determines the magnitude of the reconstruction coefficients. Thus, it requires much less auto-calibration signals (ACS) data and achieves higher acceleration factors. The results show that this method can increase the acceleration factor and the reconstruction speed while achieving the same level of reconstruction quality. Finally, a fast pTx pulse design method is proposed to accelerate the design speed. This method is based on the spatial domain pulse design method and can be used to accelerate similar methods. We substitute the two computational expensive matrix- vector multiplications in the conjugate gradient (CG) solver with gridding and fast Fourier transform (FFT). Theoretical and simulation results have shown that the design speed can be improved by 10 times. Meanwhile, the memory cost is reduced by 103 times. This breaks the memory burden of implementing pulse designs on GPU which enables further accelerations.