Dissertation Defense Announcements

The development of in-memory computing has fueled the emergence of in-memory computing systems. Data explosion is also posing an unprecedented demand for memory capacity to handle the ever-growing data size. Thus, in-memory computing systems are increasingly looking inward at hybrid memory caches of under-processed data as resources to be mined. Our preliminary study finds that some existing data management strategies often trade application performance for low memory utilization, and hence can induce frequent I/O operations between memory system and storage system.

To achieve this goal, we propose to design a hybrid memory system that includes fast and relatively slow memory hardware. In order to realize a runtime system that automatically optimizes data management on hybrid memory, we will (1) propose a new shared in-memory cache layer among parallel executors that are co-hosted on the same computing node, which aims to improve the overall hit rate of data blocks; (2) develop a middleware layer built on top of existing deep learning frameworks that streamlines the support and implementation of online learning applications; (3) design a unified in-memory computing architecture with efficient data management strategy to optimize memory allocation and recycle for ML applications.

The metropolitan areas suffer more traffic, the change in travel time is very complex as it can be influenced by various factors, many of which are also unpredictable. Random forest was applied in the travel time prediction application to overcome the overfitting problem. Furthermore, the attention mechanism is implemented by developing the neural network to capture the inner relationship within the traffic data. The proposed long short memory neural network with attention mechanism method achieves its superior capability for TTP longer than 15 minutes (30 min to 60 min), overcoming the performance issue through long temporal dependency and memory blocks. To validate the accuracy and reliability of proposed models, the proposed approaches are tested using a freeway corridor in Charlotte, North Carolina, using the probe vehicle-based traffic data. Detailed information about the input variables and data preprocessing was presented. The results indicate that all proposed TTP models predicting in 15 minutes show better prediction performance over the other time horizons. A comparison with other prediction methods validates that the proposed hybrid LSTM and RF method can achieve a better prediction performance in accuracy and efficiency, proving its deployment is one of the successful solutions to critical, real-world transportation challenges.

Courts of last resort in the United States are becoming increasingly important in American politics as the number of cases, influential decisions, and controversial issues continue to rise in the states. In discussions of federalism in the United States, these critical institutions are often overlooked as a complex system, often due to substantial data limitations on the behavior and outcomes of these courts. I situate state courts of last resort as a complex adaptive system in the broader US framework. I then seek to redress the data shortcomings by introducing a comprehensive database on state courts of last resort from 1953-2010. Using advanced data capture techniques, I evaluate my parsers to capture the ever-changing structures of the source documents. This database will be the largest in scope and case detail to date. Moreover, it should further our understanding of judicial decision making and assist the prediction of the impact of institutional change on the system.
In addition, I modeled and analyzed the system as a complex adaptive system. Since the system has network characteristics, I used the approach of network science to model the system based on the citation behavior. Moreover, I created an automated dictionary-based classification model to extract and classify the citation treatments for the court cases.
Using state-of-the-art algorithms in network science and natural language processing, I was able to analyze the system and test the performance of the algorithms based on the system characteristics.

Renewable energy-based electric power generation is receiving more attention due to the increasing power demand and environmental concerns. The interconnection of these distributed resources to the grid is based on the grid code standards to ensure power quality, reliability, and security. The intermittent nature of the renewable penetrated grid, along with phasing out of conventional generation units, new HVDC lines, and long-distance transmissions from remote areas, impose several challenges to grid stability. Hence it is crucial to explore control approaches that can efficiently control these Distributed Energy Resources (DERs) to improve the overall power quality and reliability. This dissertation presents modeling, stability studies, and advanced control architectures that can support and coordinate the Wind Energy Conversion Systems (WECS) and other inverter-based Distributed Energy Resources (DERs) to improve the quality of the generation, transmission, and distribution systems. The first part of the work proposes advanced adaptive-based robust sensorless control approaches for rotor side and grid side control of DFIG based WECS.
Further, the dissertation discusses the challenges of transferring high power of renewable penetrated grid and some possible solutions and control approaches. Finally, the work efficiently coordinates the available resources to ensure power quality in a distribution network. All the proposed designs are validated using simulation results developed by dynamic models or through real-time simulators, which proves the ability of the advanced controllers to improve grid reliability. The quantification based on standard metrics used for performance improvements discussed in each design shows that the designs have exceptional advantages compared with conventional controllers.