This thesis presents a novel approach to type theory called “pure subtype systems”, and a core calculus called DEEP which is based on that approach. DEEP is capable of modeling a number of interesting language techniques that have been proposed in the literature, including mixin modules, virtual classes, feature-oriented programming, and partial evaluation. The design of DEEP was motivated by two well-known problems: “the expression problem”, and “the tag elimination problem.” The expression problem is concerned with the design of an interpreter that is extensible, and requires an advanced module system. The tag elimination problem is concerned with the design of an interpreter that is efficient, and requires an advanced partial evaluator. We present a solution in DEEP that solves both problems simultaneously, which has never been done before. These two problems serve as an “acid test” for advanced type theories, because they make heavy demands on the static type system. Our solution in DEEP makes use of the following capabilities. (1) Virtual types are type definitions within a module that can be extended by clients of the module. (2) Type definitions may be mutually recursive. (3) Higher-order subtyping and bounded quantification are used to represent partial information about types. (4) Dependent types and singleton types provide increased type precision. The combination of recursive types, virtual types, dependent types, higher-order subtyping, and bounded quantification is highly non-trivial. We introduce “pure subtype systems” as a way of managing this complexity. Pure subtype systems eliminate the distinction between types and objects; every term can behave as either a type or an object depending on context. A subtype relation is defined over all terms, and subtyping, rather than typing, forms the basis of the theory. We show that higher-order subtyping is strong enough to completely subsume the traditional type relation, and we provide practical algorithms for type checking and for finding minimal types. The cost of using pure subtype systems lies in the complexity of the meta-theory. Unfortunately, we are unable to establish some basic meta-theoretic properties, such as type safety and transitivity elimination, although we have made some progress towards these goals. We formulate the subtype relation as an abstract reduction system, and we show that the type theory is sound if the reduction system is confluent. We can prove that reductions are locally confluent, but a proof of global confluence remains elusive. In summary, pure subtype systems represent a new and interesting approach to type theory. This thesis describes the basic properties of pure subtype systems, and provides concrete examples of how they can be applied. The Deep calculus demonstrates that our approach has a number of real-world practical applications in areas that have proved to be quite difficult for traditional type theories to handle. However, the ultimate soundness of the technique remains an open question.

This project aimed to constrain the interpretation uncertainties associated with reflection seismic data of deep-water slope deposits.  The basic premise of the project is that seismic data is affected by small-scale architectural elements and even conventional low-frequency data may contain clues of the sub-seismic geometries.  These can be decoded by understanding the interaction between internal elements and the seismic wavelet.  A series of outcrop-derived forward seismic models was created, representing different types of outcrop based slope deposits.  The seismic interpretation of the forward models was based on amplitude analysis supplemented by instantaneous attributes and spectral decomposition. In order to create realistic synthetic seismograms, input models included geometries whose thickness was as low as 1% of the resolution limit.  By revealing the influence of small-scale structures on synthetic seismic data at the high end of the spectrum (70Hz to 100Hz), the knowledge of tuning effects and the interaction between interfering reflections at lower frequencies (i.e. 20, 40 and 60Hz) could potentially be significantly improved. The gained experience was then applied to real seismic data.  It was proven that small-scale geometries have an additional, highly significant effect on the composite reflection. Because of the inherent non-uniqueness in seismic reflection, the specific seismic forward models of particular outcrop analogues can only be used as guides to the seismic interpretation of the particular architectural elements of a subsurface deposit and not as definite models against which one can definitely pattern match real and modelled seismic data.  as burial depth increases, so does the non-uniqueness of the seismic interpretation of seismic data from deposits whose internal geometries are around/below the tuning thickness.

<p>Deep brain stimulation (DBS) is an effective therapy for motor symptoms in Parkinson's disease (PD). DBS efficacy depends on the stimulation parameters, and the current gold standard therapy is high-frequency stimulation (>100 Hz) with constant interpulse intervals and short pulse widths (<210 &#956;s). However, the temporal pattern of stimulation is a novel parameter dimension that has not been thoroughly explored. We used non-regular temporal patterns of DBS to pursue two goals: to better understand the mechanisms of DBS, and to increase the efficacy and efficiency of DBS for PD.</p><p>First, we designed high frequency patterns of non-regular stimulation with distinct features proposed to be important for efficacy and evaluated these patterns in human subjects with PD. Unexpectedly, some non-regular patterns of stimulation improved performance of an alternating finger-tapping task-a proxy for bradykinesia-compared to high frequency regular stimulation. Performance in the motor task was correlated with suppression of beta band power in a computational model of the basal ganglia suggesting a possible mechanism for effective stimulation patterns.</p><p>Inspired by the increased clinical efficacy of non-regular patterns of stimulation with high average frequencies, we developed a non-regular pattern of stimulation that reduced motor symptoms in PD using a low average stimulation frequency. Since the number of potential combinations of interpulse intervals is exceedingly large and it is unclear how such timing should be selected, we applied computational evolution to design an optimal temporal pattern of deep brain stimulation to treat the symptoms of PD. Next, we demonstrated the efficacy of the resulting pattern of stimulation in hemi-parkinsonian rats and humans with PD. Both the optimized stimulation pattern and high frequency stimulation suppressed abnormal oscillatory activity in the basal ganglia in the rat and human, providing a shared mechanism of action for effective stimulation patterns. This innovation could allow patients to achieve battery life savings compared to traditional high frequency stimulation, which will reduce the costs and risks of frequent battery replacement procedures. Further, our approach can be used to design novel temporal patterns of stimulation in other applications of neural stimulation.</p><p>Finally, we explored evoked field potentials in the subthalamic nucleus (STN) in response to DBS. These potentials were evoked by stimulation through one of the contacts on the DBS lead and recorded from the two surrounding contacts. Subthalamic DBS local evoked potentials (DLEPs) have never before been recorded. We characterized the DLEPs, differences across DBS frequencies and time, their relationship to beta frequency oscillations and phase-amplitude coupling, and their dependence on electrode contact location.</p><p>A 3-dimensional biophysical model of DBS in the subthalamic nucleus-globus pallidus externus (GPe) subcircuit was built to explore the neural origin of the DLEPs. The computational model could reproduce the DLEP signal, and it revealed that the quasi-periodic DLEP oscillations are caused by excitatory synaptic currents in STN interrupted periodically by inhibition from GPe.</p><p>DLEP power was correlated with beta band oscillation power in the recordings without DBS, and significant phase-amplitude coupling was observed in a subset of subjects with robust DLEP responses. Together, all available evidence suggested the contact location was an important determinant for the presence and characteristics of DLEP signals. Predictions were made concerning contact location relative to the boundaries of the STN based on the DLEP recordings and insights gained using the computational model, and the predictions were in agreement with blinded post hoc imaging based contact localization for ~70% of contacts predicted to be within STN.</p><p>DLEPs are an exciting new signal with several useful applications. DLEPs could help neurosurgeons verify accurate DBS lead placement or optimal stimulation parameters, probe the pathological basal ganglia, and elucidate the mechanisms of DBS.</p> / Dissertation

Deep learning networks in the literature traditionally only used a single input modality (or data stream). Integrating multiple modalities into deep learning networks with the goal of correlating extracted features was a major issue. Traditional methods involved treating each modality separately and then writing custom code to combine the extracted features. Current solutions for small numbers of modalities (three or less) showed there are multiple architectures for modality integration. With an increase in the number of modalities, the “curse of dimensionality” affects the performance of the system. The research showed current methods for larger scale integrations required separate, custom created modules with another integration layer outside the deep learning network. These current solutions do not scale well nor provide good generalized performance. This research report studied architectures using multiple modalities and the creation of a scalable and efficient architecture.

The purpose of the research study was to provide improved relaxation training with the use of an electromyography feedback device based on the design of Green et al. (1969). It was intended that this instrument would allow the training of deep muscle relaxation to the point of neuro-muscular silence, while remaining inexpensive enough to be applied in the clinical setting.

The last decade has seen considerable research focusing on understanding the factors underlying schizophrenia and bipolar disorder. A major challenge encountered in studying these disorders, however, has been the contribution of genetic, or etiological, heterogeneity to the so-called “missing heritability” [1-6]. Further, recent successes of large-scale genome-wide association studies (GWAS) have nonetheless seen only limited advancements in the delineation of the specific roles of implicated genes in disease pathophysiology. The study of microRNAs (miRNAs), given their ability to alter the transcription of hundreds of targeted genes, has the potential to expand our understanding of how certain genes relate to schizophrenia and bipolar disorder. Indeed, the strongest finding of one recent mega-analysis by the Psychiatric GWAS consortium (PGC) was for a miRNA, though little can be said presently about its particular role in the etiologies of schizophrenia and bipolar disorder [52]. Next generation sequencing (NGS) is a versatile technology that can be used to directly sequence either DNA or RNA, thus providing valuable information on variation in the genome and in the transcriptome. A variation of NGS, MicroSeq, focuses on small RNAs and can be used to detect novel, as well as known, miRNAs [26,125, 126]. The following thesis describes the role of miRNAs in schizophrenia and bipolar disorder in various experimental settings. As an index of the interaction between multiple genes and between the genome and the environment, miRNAs are great potential biomarkers for complex disorders such as schizophrenia and bipolar disorder.

ABSTRACT: Whiplash-associated disorders are a common occurrence. Physiotherapy rehabilitation of such disorders include, among others, improving the recruitment ability of the deep cervical flexor muscles. Cervical proprioception, which has recently gained attention, is not commonly addressed. Evidence points to a possible link between cervical proprioception and deep cervical flexor recruitment ability. This study aimed to determine whether such a correlation exists. This is significant as it highlights the role that recruitment training of the deep cervical flexors has on cervical proprioception. A correlation study design was employed that involved 29 patients with whiplashassociated disorders. They were tested in their ability to perform the cranio-cervical flexion test and Revel’s test for proprioception. Linear regression was employed to interpret the results. This study concluded that a correlation exists between the ability to perform the craniocervical- flexion test and cervical proprioception.

Automatic target recognition capabilities in autonomous underwater vehicles has been a daunting task, largely due to the noisy nature of sonar imagery and due to the lack of publicly available sonar data. Machine learning techniques have made great strides in tackling this feat, although not much research has been done regarding deep learning techniques for side-scan sonar imagery. Here, a state-of-the-art deep learning object detection method is adapted for side-scan sonar imagery, with results supporting a simple yet robust method to detect objects/anomalies along the seabed. A systematic procedure was employed in transfer learning a pre-trained convolutional neural network in order to learn the pixel-intensity based features of seafloor anomalies in sonar images. Using this process, newly trained convolutional neural network models were produced using relatively small training datasets and tested to show reasonably accurate anomaly detection and classification with little to no false alarms. / Includes bibliography. / Thesis (M.S.)--Florida Atlantic University, 2018. / FAU Electronic Theses and Dissertations Collection

This research is describing a novel method to generate realism improved synthetic images while preserving annotation information and the eye gaze direction. Furthermore, it describes how the perceptual loss can be utilized while introducing basic features and techniques from adversarial networks for better results.

Intelligent vehicle technologies are growing rapidly that can enhance road safety, improve transport efficiency, and aid driver operations through sensors and intelligence. Advanced driver assistance system (ADAS) is a common platform of intelligent vehicle technologies. Many sensors like LiDAR, radar, cameras have been deployed on intelligent vehicles. Among these sensors, optical cameras are most widely used due to their low costs and easy installation. However, most computer vision algorithms are complicated and computationally slow, making them difficult to be deployed on power constraint systems. This dissertation investigates several mainstream ADAS applications, and proposes corresponding efficient digital circuits implementations for these applications. This dissertation presents three ways of software / hardware algorithm division for three ADAS applications: lane detection, traffic sign classification, and traffic light detection. Using FPGA to offload critical parts of the algorithm, the entire computer vision system is able to run in real time while maintaining a low power consumption and a high detection rate. Catching up with the advent of deep learning in the field of computer vision, we also present two deep learning based hardware implementations on application specific integrated circuits (ASIC) to achieve even lower power consumption and higher accuracy. The real time lane detection system is implemented on Xilinx Zynq platform, which has a dual core ARM processor and FPGA fabric. The Xilinx Zynq platform integrates the software programmability of an ARM processor with the hardware programmability of an FPGA. For the lane detection task, the FPGA handles the majority of the task: region-of-interest extraction, edge detection, image binarization, and hough transform. After then, the ARM processor takes in hough transform results and highlights lanes using the hough peaks algorithm. The entire system is able to process 1080P video stream at a constant speed of 69.4 frames per second, realizing real time capability. An efficient system-on-chip (SOC) design which classifies up to 48 traffic signs in real time is presented in this dissertation. The traditional histogram of oriented gradients (HoG) and support vector machine (SVM) are proven to be very effective on traffic sign classification with an average accuracy rate of 93.77%. For traffic sign classification, the biggest challenge comes from the low execution efficiency of the HoG on embedded processors. By dividing the HoG algorithm into three fully pipelined stages, as well as leveraging extra on-chip memory to store intermediate results, we successfully achieved a throughput of 115.7 frames per second at 1080P resolution. The proposed generic HoG hardware implementation could also be used as an individual IP core by other computer vision systems. A real time traffic signal detection system is implemented to present an efficient hardware implementation of the traditional grass-fire blob detection. The traditional grass-fire blob detection method iterates the input image multiple times to calculate connected blobs. In digital circuits, five extra on-chip block memories are utilized to save intermediate results. By using additional memories, all connected blob information could be obtained through one-pass image traverse. The proposed hardware friendly blob detection can run at 72.4 frames per second with 1080P video input. Applying HoG + SVM as feature extractor and classifier, 92.11% recall rate and 99.29% precision rate are obtained on red lights, and 94.44% recall rate and 98.27% precision rate on green lights. Nowadays, convolutional neural network (CNN) is revolutionizing computer vision due to learnable layer by layer feature extraction. However, when coming into inference, CNNs are usually slow to train and slow to execute. In this dissertation, we studied the implementation of principal component analysis based network (PCANet), which strikes a balance between algorithm robustness and computational complexity. Compared to a regular CNN, the PCANet only needs one iteration training, and typically at most has a few tens convolutions on a single layer. Compared to hand-crafted features extraction methods, the PCANet algorithm well reflects the variance in the training dataset and can better adapt to difficult conditions. The PCANet algorithm achieves accuracy rates of 96.8% and 93.1% on road marking detection and traffic light detection, respectively. Implementing in Synopsys 32nm process technology, the proposed chip can classify 724,743 32-by-32 image candidates in one second, with only 0.5 watt power consumption. In this dissertation, binary neural network (BNN) is adopted as a potential detector for intelligent vehicles. The BNN constrains all activations and weights to be +1 or -1. Compared to a CNN with the same network configuration, the BNN achieves 50 times better resource usage with only 1% - 2% accuracy loss. Taking car detection and pedestrian detection as examples, the BNN achieves an average accuracy rate of over 95%. Furthermore, a BNN accelerator implemented in Synopsys 32nm process technology is presented in our work. The elastic architecture of the BNN accelerator makes it able to process any number of convolutional layers with high throughput. The BNN accelerator only consumes 0.6 watt and doesn't rely on external memory for storage.