Upcoming Doctoral Dissertations

" The Spin-Orbit Torque Magnetic Random-Access Memory (SOT-MRAM) is a type of non-volatile memory that uses SOT to write data instead of traditional magnetic fields or electrical currents. This technology is promising for future memory applications due to its potential for lower power consumption, faster operation, and scalability compared to other types of MRAM. Heavy metals exhibiting strong spin-orbit coupling and topological insulators with unique surface states that support spin-momentum locking can generate substantial spin currents when subjected to an electrical current. When these materials are paired with van der Waals (vdW) or non-vdW magnets, high-efficiency spin-torque transfer to the magnetic layers can be achieved. The strong perpendicular magnetic anisotropy (PMA) in vdW magnets is particularly beneficial for spintronics applications." Read more...

" Optical coherence tomography (OCT) is a biomedical imaging modality which offers depth-resolved, realtime, and high-resolution scanning. Optical coherence tomography systems can achieve a penetration depth of up to 10 mm, a video-rate data acquisition, and a high axial resolution at micron level. However, it is challenging to simultaneously maintain high scanning rate and image quality for OCT systems. Moreover, pathological information, which is essential for clinical decision making, can not be directly obtained during OCT imaging process. " Read more...

" Recent advances in machine learning have yielded remarkable gains across vision, language, and graph-based tasks. Yet, persistent vulnerabilities to adversarial perturbations raise significant challenges for deploying these systems in safety-critical domains. This dissertation tackles the robustness problem from a unified, spectral perspective on manifold learning. It develops near-linear-time frameworks that analyze and enhance stability by studying how input–output distance mappings become distorted under learned models. First, SPADE (Spectral Method for Black-Box Adversarial Robustness Evaluation) addresses general deep neural networks. SPADE constructs low-dimensional input and output manifolds and uses effective resistance distances to detect how small input changes can trigger large output distortions. This yields an upper bound on the model’s Lipschitz constant without white-box model access. Empirical tests on MNIST and CIFAR-10 confirm that SPADE pinpoints highly fragile data samples and strengthens adversarial training." Read more...

" The links between nutrition and health are well documented, connecting foods and nutrients to long and short-term health outcomes. Lacking, however, are data-driven approaches that take advantage of advancements in machine learning, a particularly useful endeavor given the complex and multi-faceted nature of nutrition. In this thesis we look at how data-driven approaches to nutritional epidemiology and dietary assessment can offer a new perspective from which to view our diets, particularly during pregnancy, an under-researched stage of life." Read more...

" Machine learning (ML) algorithms are applied across various domains. Despite their success, concerns about privacy leakage remain a significant challenge. In this dissertation, I focus on the disparities in privacy risks associated with ML algorithms. I specifically concentrate on Membership Inference Attacks (MIAs), and I demonstrate that privacy leakage varies at both the sample level and the model level. " Read more...

" Over the past decade, lithium-ion batteries have evolved into indispensable energy sources for portable devices, electric vehicles, and large-scale energy storage systems. Currently, commercially available batteries predominantly utilize a limited set of cathode types composed of transition metals (TMs) and lithium (Li) frameworks. However, even as researchers continually refine these layered frameworks, they often suffer from irreversible phase transitions, oxygen release, and lattice distortion under prolonged cycling and at extremely high voltage." Read more...

" Quantum technologies rely on robust single photon sources, whose emission can be precisely controlled for use in secure quantum communication, sensing and scalable on-chip photonic circuitry. Achieving such control typically requires quantum emitters (QEs) that feature narrow spectral linewidth and high brightness. Equally important is the ability of dynamic high-speed manipulation of these QEs to tune their emission energies on demand for photonic circuit integration to enable on-chip technologies. While 2D semiconductors such as hexagonal boron nitride (hBN) and transition metal dichalcogenides (TMDs) have proven fertile ground for discovering new quantum light sources, scalable fabrication of QEs with desired properties for quantum information technology and integrating them with advanced tuning strategies remains a critical challenge as being addressed in my work. " Read more...

" The complex nature of flow and transport in porous media has led to substantial research across various disciplines in recent decades. One of the key porous media properties that controls fluid flow is permeability. In subsurface environments, the spatial variability of permeability, along with the existence of natural fractures and anisotropic properties of the media, result in highly unpredictable flow patterns. A comprehensive investigation of permeability across scales— from pore to field scale —is essential for enhancing fluid behavior models in applications such as carbon sequestration, groundwater management, and energy storage. " Read more...

" Sustainable product design requires new perspectives and methods to achieve environmental, social, and economic success. The multi-agent, multi-variable, complex environment in which stakeholders of a product coexist creates a complex system, which is known as the wicked problem of sustainability. Sustainable product design is an approach to addressing sustainability challenges through product development processes and tools. The number of sustainable design methods has been increasing rapidly in recent years, but their adoption is limited, and many of these methods exclusively focus on the environmental impacts of products rather than taking a holistic perspective that includes social and economic sustainability. " Read more...

" Since its inception in the early 1960s, Light Detection and Ranging (LiDAR) has rapidly advanced, demonstrating its capabilities across various fields. On the other hand, elastography is an important tool for biomedical diagnoses to differentiate between healthy and cancerous tissue. To date, no research has combined LiDAR technology with elastography, and this dissertation presents the first successful attempt to explore this uncharted field. " Read more...

" Given the escalating impacts of climate change on coastal communities worldwide, there is a pressing need for enhancing coastal flood mitigation strategies in the face of increasing storm intensities and sea-level rise. Traditional approaches to coastal defense, typically reliant on hard infrastructure, have shown limitations in terms of environmental impact, sustainability, and adaptability to changing conditions. These limitations underscore the necessity for innovative solutions that integrate the resilience of natural systems with the protective certainty of engineered structures. However, a significant gap exists in quantitatively assessing the resilience of such hybrid systems during extreme events and in designing optimized characteristics considering the economic aspects of coastal defense projects. " Read more...