Upcoming Doctoral Dissertations

"Visual-Inertial Simultaneous Localization and Mapping (VI-SLAM) systems encounter considerable challenges in demanding environments, making advancements in this area crucial. Integrating a single camera with an Inertial Measurement Unit (IMU) in Visual-Inertial Navigation Systems (VINS) offers a cost-effective, low-power solution suitable for robot perception and AR/VR applications. The camera captures rich environmental details, while the IMU measures acceleration and angular velocity, enhancing system resilience in fast-motion or low-texture scenarios. This sensor fusion allows for complementary benefits, though achieving precise and dense 3D reconstruction in challenging conditions remains an unresolved issue." Read more...

"Camera pose estimation aims to localize an observer w.r.t. a known geometric reference and has a range of applications from autonomous navigation to virtual/augmented reality. Motivated by the ubiquity of digital cameras deploying rolling shutter (RS) hardware, pose estimation modules have been extended from the pinhole camera model to incorporate the dynamic 1D capture characteristics of these sensors. However, absolute pose estimation from single RS scanline inputs has been hitherto ignored in the literature. This omission may be attributed to the limited geometric context available for instantaneous RS capture and stringent latency requirements involved in real-time system operation. " Read more...

"Understanding complex neuronal firing patterns and interactions between neural circuits at different brain regions is essential for uncovering the mechanisms of brain function and dysfunctions. Electrophysiological Source Imaging (ESI) refers to reconstructing underlying cortical and subcortical electrical activities from electroencephalography (EEG) or magnetoencephalography (MEG) recordings. ESI is crucial for both neuroscience research and clinical applications, serving as an essential tool for capturing brain source signals with high temporal resolution. However, solving the ESI inverse problem remains challenging due to its illposed nature. To obtain a unique solution, traditional algorithms emphasize incorporating predesigned neurophysiological priors to restrict the solution space, while deep learning frameworks aim to directly learn the mapping from scalp EEG/MEG measurements to underlying brain sources in a data-driven manner, eliminating the need for handcrafted priors. Building on this, this dissertation proposes several ESI algorithms designed to capture the complexities of brain activity by leveraging sparse and graph-structured EEG/MEG signals, especially in cases involving extended source activities." Read more...

"Recent advancements in thin-film lithium niobate technology have opened new avenues for creating high-performance photonic devices. Thin-film lithium niobate combines the exceptional qualities of bulk lithium niobate with the advantages of a thin-film format, including improved light confinement, reduced propagation losses, and enhanced material interactions. These features are essential for achieving high-efficiency frequency conversion and precise dispersion engineering in integrated photonic platforms." Read more...

"Adequate signal to noise is a prerequisite for quantum communications, computation, sensing, and imaging. To this end, high dimensional signal encoding can maximize channel and bandwidth utilization for high data rates. Extracting information from photon-starved and noisy environments allows long-distance ranging, sensing, and imaging. This dissertation presents low-noise quantum frequency conversion (QFC) and its various implementations using lithium-niobate crystals, waveguides, and nanophotonic circuits, and explores their applications in sensing and communications. " Read more...

"This dissertation explores innovative approaches to enhancing hydrological modeling by addressing critical challenges in flood inundation mapping (FIM) and water resources management practices. The initial phase involves developing a high-resolution FIM framework based on the Weather Research and Forecasting Model Hydrological modeling system (WRF-Hydro), the computational engine of the National Water Model (NWM), and a conceptual FIM method called Height Above the Nearest Drainage (HAND). This framework was designed to support an existing regional flood advisory system using lumped hydrological modeling approach for urban areas. " Read more...

"Typically, operation tasks within the power system field follow a predict-then-optimize framework, in which machine learning (ML) methods are first trained to predict key parameters and then optimization models use these predictions as inputs to determine optimal operational decisions. For instance, renewable energy availability is predicted to serve as inputs for day-ahead operation models. The ultimate goal of such a predict-then-optimize process is to achieve the best operation economics associated with the optimal operation tasks, e.g., minimum operation cost or maximum operation revenue." Read more...

"Robots are increasingly becoming integral parts of our daily lives. Achieving safe and efficient navigation in complex and dynamic environments shared with humans presents significant challenges. This dissertation addresses the challenges of autonomous navigation in dynamic environments by developing learning-based methods that enable robots to navigate collision-free and to respond to natural language instructions. Traditional navigation systems fall short in these scenarios due to their inability to capture human social behaviors. By leveraging human trajectory data and advanced robotic simulation techniques, this research provides innovative solutions to improve robot navigation and human-robot interaction." Read more...

"The three essays in my dissertation proposal examine the use of machine learning and artificial intelligence (ML/AI) for performance improvement in large-scale combinatorial problems and time series forecasting. The first essay, “Using ML/AI to Improve Computational Performance in Large-scale Optimization Problems” surveys recent research on the use of ML/AI techniques to improve computational performance in large-scale combinatorial optimization (CO) problems. These problems may be NP-hard or beyond. Exact and heuristic optimization algorithms have been designed to solve CO problems. However, many CO problems cannot be solved in a reasonable time by traditional approaches. I survey research on ML/AI approaches to this problem. I develop a framework for categorizing the various approaches based on whether they involve algorithm imitation or algorithm generation. The survey concludes that ML/AI approaches have promise but that a general approach to solving a broad class of CO problems has yet to be discovered." Read more...