Committee:
Dr. David Citrin, ECE, Chair, Advisor
Dr. David Anderson, ECE
Dr. Ryan Sherman, CEE
Dr. Doug Yoder, ECE
Dr. Nico Declercq, ME
]]>Committee:
Dr. Milos Prvulovic, CoC, Chair, Advisor
Dr. David Anderson, ECE
Dr. Matthieu Bloch, ECE
Dr. Gregory Durgin, ECE
Dr. Celine Lin, CoC
]]>Date: May 21st, 2026
Time: 3:00 PM – 4:30 PM EST
Location: Groseclose 404 and Zoom
Meeting Link: https://gatech.zoom.us/j/97324269229
Caleb Ju
Ph.D. Candidate in Operations Research
School of Industrial and Systems Engineering
Georgia Institute of Technology
Committee:
Dr. Guanghui Lan (advisor), School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Yuejie Chi, Department of Statistics and Data Science, Yale University
Dr. Constance Crozier, School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Katya Scheinberg, School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Alexander Shapiro, School of Industrial and Systems Engineering, Georgia Institute of Technology
Abstract:
This thesis focuses on the design and implementation of stochastic optimization methods towards dynamic decision-making under uncertainty. This includes problems such as reinforcement learning (RL), multi-stage stochastic programs, and stochastic optimal control, as well as applications in energy and sustainability. A central theme is developing new algorithms with state-of-the-art sample complexity under relaxed assumptions that better match practice.
This thesis starts by deriving new convergence guarantees and termination criteria for finite state and action Markov decision processes (MDPs) and RL problems. Chapter 2 introduces a new advantage gap function for these problems. This gap function provides close approximations of the (unknown) optimality gap, which can be easily estimated in a data-driven manner for RL. Moreover, by incorporating the gap function into the design of step size rules, we demonstrate that policy gradient methods can solve MDPs in strongly-polynomial time. This result shows popular gradient-based approaches can efficiently find exact solutions when the model is known, and it matches the strongly-polynomial runtime of simplex and Howard’s policy iteration proven by Ye. In Chapter 3, we develop a novel framework called auto-exploration for solving RL problems in the online (or single-trajectory) model. We use the framework to derive a new algorithm-independent sample complexity under weaker mixing assumptions on the optimal policy. Moreover, our algorithm is parameter-free since it does not require a priori knowledge of the unknown mixing time. Additionally, the method can easily incorporate linear function approximation.
After investigating finite state and action problems, the thesis advances towards RL and classical control problems over continuous spaces. In Chapter 4, we revisit the linear quadratic regulator in the online model. Despite the non-convexity of the problem in policy space, we design a globally convergent natural policy gradient paired with a new conditional stochastic primal-dual algorithm. This combined algorithm delivers state-of-the-art sample complexity under a relaxed assumption on the stability of the initial controller (rather than the stability of all intermediate controllers, which is commonly posited in prior art). Then Chapter 5 introduces a new policy dual averaging (PDA) for solving RL problems over general state and action space. PDA can easily incorporate function approximation (e.g., nonlinear kernels and neural networks) while providing efficient global convergence guarantees. Preliminary numerical results demonstrate the robustness of PDA and show it can be competitive with state-of-the-art RL algorithms. In Chapter 6, we study stationary stochastic programs over an infinite horizon. We introduce a continually-exploring infinite-horizon explorative dual dynamic programming. Compared to the celebrated stochastic dual dynamic programming, our method explores the feasible region longer and updates the cutting-plane model more frequently. These innovations yield new iteration complexities while offering numerically efficient performance on inventory control and hydrothermal planning problems.
The thesis concludes with RL applications towards energy and sustainability. Chapter 7 applies RL towards the operation of grid-scale batteries co-located with solar generation. Compared to simpler rules-based control, we show RL has two significant downstream effects: (1) solar energy is more effectively shifted towards high demand periods, and (2) potential to reduce ramping issues caused by super-position of many similar battery operations. In Chapter 8, we utilize RL for real-time control of multiple feedstocks in waste biorefining processes. In the short term, RL achieves faster target tracking with increased precision and accuracy, while in the long term, it shows adaptive and robust behavior even under additional seasonal supply variability, meeting downstream demand with high probability.
]]>
Date: Friday May 15th, 2026
Time: 1:00pm - 2:00pm ET
Location: Guggenheim 244
Virtual: https://teams.microsoft.com/meet/295530221364067?p=XAsX2uMHAJ40THTLPb
Johnathan Corbin
Robotics Ph.D. Student
Daniel Guggenheim School of Aerospace Engineering
Georgia Institute of Technology
Committee:
Dr. Jonathan Rogers (Advisor)
Daniel Guggenheim School of Aerospace Engineering
Georgia Institute of Technology
Dr. Sarah H. Q. Li
Daniel Guggenheim School of Aerospace Engineering
Georgia Institute of Technology
Dr. Anirban Mazumdar
George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Dr. Matthew Hale
School of Electrical and Computer Engineering
Georgia Institute of Technology
Dr. Sean Wilson
Robotics and Autonomous Systems Division
Georgia Tech Research Institute
Abstract:
In heterogeneous multi-agent systems, ensuring safety requires two distinct decisions. The first is determining what corrective action is needed. The second is assigning which agents should take it. Standard control barrier function (CBF) formulations combine these into a single centralized quadratic program, distributing corrective effort by geometric proximity with no regard for agent capability, privacy, or accumulated burden. This can cause actuator-limited agents to be unable to maintain safety and makes it difficult to adjust how responsibility is allocated to agents.
This proposal reformulates multi-agent safety-critical control as a feasible resource allocation problem through a two-stage architecture. The first stage compresses multi-constraint safety into a single, dynamically feasible "safety deficit" using integral augmentation, log-sum-exponential composition, and optimal-decay CBFs. The second stage introduces "avoidance credit," an allocable resource that decouples the physics of safety from the economics of effort distribution through a modular interface. The interface admits any allocation mechanism satisfying a small set of conditions, enabling properties such as agent privacy and long-horizon fairness. As long as the allocation satisfies these conditions, the system is guaranteed to maintain safety within the actuator limits of the agents.
]]>
Date: Wednesday, May 20th, 2026
Time: 1pm - 2pm ET
Location: MRDC 3515 (https://teams.microsoft.com/meet/225122560859133?p=BordRn7dHFBulSff2T)
Seth Golembeski
Robotics Ph.D. Student
Guggenheim School of Aerospace Engineering
Georgia Institute of Technology
Committee:
Dr Anirban Mazumdar (Advisor): School of Mechanical Engineering
Dr Shreyas Kousik: School of Mechanical Engineering
Dr Jonathan Rogers: School of Aerospace Engineering
Dr Kyriakos Vamvoudakis: School of Aerospace Engineering
Dr Scott Nivison: Air Force Research Lab
Abstract:
Many robotics problems involve multiple interacting agents, often with conflicting objectives. Robust control addresses problems involving systems that reject external disturbances, such as wind, uneven terrain, or mechanical vibrations. Adversarial settings consider agents that are in opposition, such as combat, markets, or navigating a crowd.
These seemingly disparate problems are connected by Nash equilibria: the combination of all agents’ actions from which no agent can unilaterally deviate to improve its reward. These solutions are not always globally optimal, but each agent can guarantee that it is locally optimal with respect to unilateral deviations by others. Computing Nash equilibria and optimal control solutions are exceptionally difficult. Most existing methods assume monotonicity and continuity to guarantee convergence – conditions rarely met in robotics systems, which are often complex, discontinuous, and operate in high-dimensional spaces.
Model-Based Diffusion alleviates the above problems, but is restricted to single-agent, non-game-theoretic problems, lacks formal convergence analysis, and requires large numbers of samples. To this end, this work proposes 1) MBD efficiency improvements via adaptive importance sampling 2) a decentralized, latency-robust, multi-agent diffusion algorithm for cooperative (potential) games and 3) a fully game-theoretic version of MBD for adversarial games and robust control, with convergence analysis.
]]>
I cordially invite you to attend my dissertation defense scheduled for Thursday, May 21st, from 10:00 to 11:30 AM in Room 223, Scheller College of Business. Please find an overview of the dissertation included below. Copies will be made available upon request.
You are also welcome to join virtually via the following Zoom link: https://gatech.zoom.us/j/91943023069
Kind regards,
Eun Soo Son
PhD Candidate, Organizational Behavior
Georgia Institute of Technology | Scheller College of Business
eunsoo.son@scheller.gatech.edu
Area: Organizational Behavior
Committee Members: Dr. Christina E. Shalley (Chair), Dr. Terry C. Blum, Dr. Katie L. Badura, Dr. Hyunsun Park, and Dr. Fadel K. Matta (University of Georgia)
Title: Waking Up from a Daydream: Unveiling the Curiosity-Driven Outcomes of Social Daydreaming at Work
Dissertation Overview:
Daydreaming—the act of mind wandering that departs from the here and now—occupies a substantial proportion of waking thought and much of it is known to be predominantly social. While daydreaming of any kind has long been viewed as futile, recent research highlights the affective and relational benefits of social daydreaming in particular. Acknowledging its prevalence and extending existing findings, social daydreaming may act as an underrated driver of employee outcomes in the workplace, especially by engaging thoughts and images that vary in psychological distance. Hence, drawing on construal level theory, I propose that daydreaming involving others elicits I-type and D-type curiosity—the desire for new information or knowledge to either explore new areas or resolve uncertainty—in distinct ways, with the construal level during social daydreaming shaping how each type arises. I further argue that these two types of curiosity, prompted by social daydreams, differentially motivate workplace behaviors, as I-type curiosity drives exploration while D-type curiosity fosters exploitation. Results across the two studies—an experiment using a critical incident technique and a field experience sampling study—yield mixed findings, providing partial support for the model. Overall, the findings support the effect of high construal level on I-type curiosity and the interactive effect of social daydreaming and construal level on I-type curiosity, which in turn relates to exploration and exploitation. Theoretical and practical implications of this model are discussed.
]]>
Jitesh Jain
Ph.D. Student in Computer Science
School of Interactive Computing
Georgia Institute of Technology
https://praeclarumjj3.github.io/
Date: May 22, 12:00 - 2:00 PM EST
Location: Coda 1215
Zoom: https://gatech.zoom.us/j/9814414092?pwd=WnpkTjNhRHhYQlNzZGxTTW9SWmtJdz09
Committee:
Dr. Humphrey Shi (Advisor) - School of Interactive Computing, Georgia Institute of Technology
Dr. Zsolt Kira - School of Interactive Computing, Georgia Institute of Technology
Dr. Kartik Goyal - School of Interactive Computing, Georgia Institute of Technology
Dr. Judy Hoffman - Donald Bren School of Information and Computer Sciences, University of California, Irvine
Dr. Jianwei Yang - Member of Technical Staff, xAI
Abstract: Multimodal large language models have made impressive strides in language understanding and reasoning yet struggle with abilities that come naturally to humans: perceiving objects in cluttered scenes, remembering context across long interactions, and reasoning adaptively over extended time horizons. In this thesis, we argue that overcoming this gap requires integrating three capabilities that remain weak in current systems: visual perception, multimodal memory, and any-horizon reasoning.
We begin by identifying that vision-language models fail at basic object-level perception and show that incorporating structured segmentation and depth signals as visual inputs significantly improves performance. Second, we improve spatial reasoning more fundamentally by distilling expert visual knowledge into the model's internal representations during pre-training, with no added cost at inference. Third, we build a multimodal agent with a graph-structured cognitive memory that enables efficient retrieval of multimodal context across long conversations. Finally, we propose an adaptive agent system to reason over long videos, addressing the challenges of scalable data collection, system design and training recipe for open-ended video understanding.
]]>
Date: Wednesday, May 20th, 2026
Time: 10am - 12pm ET
Location: Klaus 1212 (Zoom link)
Maxwell Asselmeier
Robotics Ph.D. Student
Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Committee:
Dr. Ye Zhao (co-advisor) – Woodruff School of Mechanical Engineering, Georgia Institute of Technology
Dr. Patricio A. Vela (co-advisor) – School of Electrical and Computer Engineering, Georgia Institute of Technology
Dr. Sehoon Ha – School of Interactive Computing, Georgia Institute of Technology
Dr. Lu Gan - Daniel Guggenheim School of Aerospace Engineering, Georgia Institute of Technology
Dr. Zak Kingston – Department of Computer Science, Purdue University
Abstract:
In this proposal, we will discuss two perception-informed and semantically-aware autonomy frameworks, one for Quadrupedal Navigation (QuadNav) and one for Bipedal Loco-manipulation (BiLocoManip). Through presenting these two frameworks, we will demonstrate how the task, environment, and embodiment at hand are crucial aspects that should inform the design of an autonomy stack. Within QuadNav, we decompose the task of navigation into global and local levels, with the local level being further decomposed into torso-, foot-, and joint-level reasoning. Torso-level decision-making is done through QuadGap, a gap-based local planner. Foot- and joint-level planning is done through QuadPiPS, a bi-level graph search and trajectory optimization framework. These various framework levels are all coordinated through a learned experience heuristic. Within BiLocoManip, we decompose the task of whole-body loco-manipulation into task planning over a long horizon, contact planning over a short horizon, and joint control in the immediate future. At the task level, a Vision Language Model (VLM) synthesizes planning commands according to environmental affordances which inform sampling during Model Predictive Path Integral (MPPI) control through a learned world model. Joint-level commands are ultimately tracked through a whole-body reinforcement learning policy.
]]>BioE PhD Proposal Presentation
9:00 am, Wednesday May 20, 2025
Location: MoSE G021
Advisor: Dr. John Blazeck (School of Chemical and Biomolecular Engineering, Georgia Institute of Technology)
Committee Members:
Dr. Mark Stycztnski (School of Chemical and Biomolecular Engineering, Georgia Institute of Technology)
Dr. Gabe Kwong (Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology & Emory University)
Dr. Alexander Vlahos (Wallace H. Coulter Department of Biomedical Engineering, Georgia Institute of Technology & Emory University)
Dr. Sarwish Rafiq (Department of Hematology and Medical Oncology at Emory University School of Medicine)
Adenosine-Inducible Self-limiting Armored CAR-T cells for Breast Cancer Treatment
Chimeric antigen receptor (CAR) T cell therapy faces challenges in solid tumors, where the immunosuppressive tumor microenvironment (TME) is a major factor that limits efficacy of CAR-T cells. Specifically, breast cancer is a type of solid tumor that has been shown dysregulated metabolism leading to buildup of immunosuppressive byproducts such as adenosine (ADO). This high ADO environment suppresses the cancer killing function of CAR-T therapy. To address the immunosuppressive environment in TME, recent clinical trials have demonstrated that cytokine-armored CAR-T cells can overcome immunosuppression through tumor-specific cytotoxicity. However, constitutive cytokine expression introduces systemic toxicities, thereby limiting therapeutic doses. Additionally, adenosine deaminase can degrade extracellular ADO into the benign inosine metabolite, thereby relieving TME immunosuppression and restoring T cell function. These findings highlight the need for a platform that combines ADO degradation with tumor specific cytokine delivery to simultaneously resolve toxicity and efficacy barriers. Therefore, this thesis proposes an ADO-inducible, self-limiting CAR-T platform that enables tumor- localized cytokine expression in response to tumor-derived ADO. Aim 1 will define the activation threshold, dynamic range, and shutdown kinetics of the ADO-inducible cytokine
platform in ADA2Ph2 secreted HEK293T cells. Aim 2 will translate the optimized circuit into primary CAR-T cells and validate tumor-localized activation in vivo. Aim 3 will evaluate
antitumor efficacy and systemic safety in mouse breast cancer models, including in combination with immune checkpoint blockade. Together, we will establish a mechanistically defined, metabolism-guided CAR-T platform that reprograms the immunosuppressive breast cancer TME while tumor localized cytokine activity, providing a broadly applicable framework for safer and more effective breast cancer immunotherapy.
]]>
Sherilyn Francis
Ph.D. Student in Human-centered Computing
School of Interactive Computing
Georgia Institute of Technology
Date: May 14, 2026
Time: 9:00 am - 12:00 pm EST
Location: online
https://gatech.zoom.us/j/95301312525
Committee
Dr. Andrea Parker (Advisor) - School of Interactive Computing, Georgia Institute of Technology
Dr. Neha Kuma - School of Interactive Computing, Georgia Institute of Technology
Dr. Naveena Karusala - School of Interactive Computing, Georgia Institute of Technology
Dr. Daprim Ogaji - African Centre of Excellence in Public Health & Toxicological Research, University of Port Harcourt
Dr. Mercy Asiedu, Google Research
Abstract
Women’s health remains structurally underrepresented in clinical research, healthcare delivery, and technology design, creating persistent inequities in diagnosis, support, and self-management across reproductive and perinatal life stages. This dissertation argues that women’s health AI should be understood not only as a technical problem, but as a sociotechnical problem of representation, interaction and trust, governance and accountabilty, and evaluation and equity. It asks how women’s health AI should represent women’s lived experiences, support safe and trustworthy interaction, distribute responsibility between patients, clinicians, and AI systems, and be evaluated in ways that demonstrate equity rather than assume it.
To address these questions, the dissertation integrates completed and proposed studies across digital health and AI. The completed empirical studies examine Black women’s postpartum health in rural Georgia and Black women’s sexual and reproductive health and HIV prevention in the United States, indicating that women engage health technologies when those systems are culturally grounded, privacy-aware, trustworthy, and responsive to lived constraints. A scoping review of participatory AI in women’s health demonstrates that participation has most often shaped interaction design and problem framing, but has rarely been translated into shared governance, model-level evaluation, or equity-demonstrating evidence. A decolonizing analysis of Black women’s sexual and reproductive health further shows that women’s health technologies must confront historical power, epistemic exclusion, and culturally narrow assumptions about whose knowledge counts.
Building on these findings, the proposed studies focus on perinatal mental health in Nigeria to derive clinician-defined interaction contracts and patient-grounded evaluation criteria for maternal health AI. Together, the dissertation contributes a methodological pathway for designing women’s health AI that is grounded in women’s definitions of harm, helpfulness, and acceptable support, and that treats equity as a property of representational design, governance, and evaluative rigor across cultural contexts.
Date: Tuesday, May 19, 2026
Time: 4PM to 6PM Eastern Time
Location: Klaus 2100
Virtual Meeting: https://gatech.zoom.us/my/li.sixu
Committee members:
Dr. Yingyan (Celine) Lin, College of Computing, Georgia Institute of Technology
Dr. Josiah Hester, College of Computing, Georgia Institute of Technology
Dr. Hyesoon Kim, College of Computing, Georgia Institute of Technology
Dr. Tushar Krishna, College of Computing, Georgia Institute of Technology
Dr. Thierry Tambe, Department of Electrical Engineering, Stanford University
Abstract:
This dissertation develops a hardware-centric specialization framework for multi-stage intelligence pipelines, using 3D intelligence as the primary study domain. Pipelines in this domain span perception and reconstruction, rendering, and high-level reasoning; their stages differ fundamentally in computational regularity, memory behavior, and control-flow dynamics, and therefore interact with current hardware to very different degrees.
We characterize this heterogeneity through algorithmic entropy, a hardware-oriented measure of execution unpredictability that decomposes into two orthogonal axes: intra-operator entropy (X, datapath irregularity) and inter-operator entropy (Y, scheduling unpredictability), quantified via GPU profiling across nine representative workloads. The resulting two-dimensional space partitions workloads into quadrants, each mapping to a substrate along a hierarchical specialization spectrum: Q1 (low X, low Y) → dedicated ASIC; Q2 (high X, low–moderate Y) → enhanced fixed-function GPU; Q3 (low X, high Y) → heterogeneous GPU-PIM; Q4 (high on both axes) → commodity GPU/CPU baseline. Three systems instantiate this principle, one per populated quadrant:
Fusion-3D (Q1): a dedicated 3D-reconstruction ASIC with hierarchical spatial tiling and a unified on-chip pipeline, extended to multi-chip for large scenes. Achieves 2.5× / 6× throughput over prior accelerators in reconstruction and inference, validated on a silicon prototype.
GauRast (Q2): a 3D Gaussian Splatting rasterizer that extends the existing GPU fixed-function rasterizer with lightweight neural-rendering operations. Achieves 6× / 4× end-to-end speedup on original / optimized 3DGS pipelines at ≤0.2% SoC area overhead.
ORCHES (Q3): a GPU-PIM heterogeneous system for test-time-compute (TTC) LLM/VLM reasoning, combining adaptive workload assignment, branch-prediction-guided pipelining, and fragmentation-aware memory structuring. Achieves 4.16× / 3.10× end-to-end speedup on text / vision reasoning over SOTA GPU baselines.
Across the three quadrants, profile-guided substrate selection delivers substantial speedups for 3D intelligence workloads while specializing only the axis where existing hardware fails. Beyond 3D intelligence, we anticipate the same (X, Y) methodology serving as a starting point for substrate selection in other multi-stage, heterogeneous-workload domains.
]]>Thesis Title: Standardized, data-driven framework for C. elegans brain atlases across development
Thesis Advisor: Dr. Hang Lu
Thesis Co-Advisor: N/A
Committee Members: Dr. Matthew Realff (School of Chemical and Biomolecular Engineering, Georgia Tech), Dr. Lily Cheung (School of Chemical and Biomolecular Engineering, Georgia Tech), Dr. Nathan McDonald (School of Biological Sciences, Georgia Tech), Dr. Yun Zhang (Organismic and Evolutionary Biology, Harvard University)
Date: 5/20/2026
Time: 2 PM
Location: EBB 3029
]]>Student Name: Jiaqi Zhang
Thesis Title: Mechanistic Studies on Degradation of Amine-Based Sorbent for Direct Air Capture (DAC)
Thesis Advisor: Christopher W. Jones
Thesis Co-Advisor: Carsten Sievers
Committee Members: David W. Flaherty (ChBE); Bjarne Kreitz (ChBE); Will Gutekunst (Chem)
Date: 5/22/2026
Time: 9am
Location: L1105 Classroom ES&T
]]>Student Name: Khanh Le
Thesis Title: Reaction Mechanisms and Intrapore Interactions of The Direct Amination of Alcohols over Zeolites
Thesis Advisor: David W. Flaherty
Thesis Co-Advisor: N/A
Committee Members: Christopher W. Jones (ChBE), William J. Koros (ChBE), Bjarne Kreitz (ChBE), Jesse G. McDaniel (School of Chemistry & Biochemistry)
Date: 5/21/2026
Time: 11:00 am - 1:00 pm
Location: EBB Krone 4029 Conference Room
]]>
BioE Ph.D. Proposal Presentation
Date and Time: Friday, May 15th, 2026, at 1:00 PM (EDT)
Location: EBB Krone Conference Room 3029
https://gatech.zoom.us/j/97249014901
Advisor: Hang Lu, Ph.D. (Chemical and Biomolecular Engineering, Georgia Institute of Technology)
Committee:
Gordon Berman, Ph.D. (Biology, Emory University)
Matthieu Bloch, Ph.D. (Electrical and Computer Engineering, Georgia Institute of Technology)
Anqi Wu, Ph.D. (Computational Science and Engineering, Georgia Institute of Technology)
Patrick McGrath, Ph.D. (Biological Sciences, Georgia Institute of Technology)
Experience-Dependent Reweighting of Conserved Neural Circuit Dynamics in C. elegans
Learning allows animals to use past experience to change how sensory cues guide future behavior, but it remains unclear how this change is implemented in brain-wide neural activity. One possibility is that learning creates a new neural response pattern; another is that learning changes how sensory cues recruit pre-existing patterns already present in the circuit. Distinguishing these possibilities requires treating neural activity as a time-evolving population process, rather than only asking which neurons increase or decrease their activity. C. elegans is well suited for this problem because controlled sensory stimulation, whole-brain calcium imaging, quantitative behavior, and genetic perturbation can be combined in the same animal. This thesis investigates how aversive olfactory learning changes brain-wide neural dynamics in C. elegans, using odors from standard bacterial food (E. coli OP50), pathogenic bacteria (Pseudomonas aeruginosa PA14), and buffer controls. Aim 1 will identify reproducible odor-evoked dynamical features across animals using latent dynamical modeling and invariant summaries. Aim 2 will test whether pathogen training shifts pathogen-evoked dynamics toward food-associated structure or produces a distinct trained-state response. Aim 3 will determine whether learning-sensitive dynamical readouts predict turning behavior and reveal how targeted genetic perturbations disrupt learned sensorimotor computation.
]]>Roshaun C. Titus
Advisor: Prof. Rosario A. Gerhardt
will defend a doctoral thesis entitled,
Processing and Microstructural Effects on the Electrical Response and Quality Assessment of Beta SiC Composites for Electronic Devices
On
Thursday, May 14 at 11:00 a.m.
LOVE Manufacturing Room 184
771 Ferst Dr NW, Atlanta, GA 30332
and/or
Virtually via MS Teams
Roshaun Titus Dissertation Defense | Teams Link
Committee
Prof. Rosario A. Gerhardt – School of Materials Science and Engineering (advisor)
Prof. Chaitanya Deo – School of Mechanical Engineering
Prof. Robert Speyer – School of Materials Science and Engineering
Prof. Preet Singh – School of Materials Science and Engineering
Dr. J. Elliot Fowler – Principal Investigator, Sandia National Laboratory
Abstract
This dissertation investigates the electromagnetic properties of cubic 3C (beta) silicon carbide (SiC) composites. SiC is a crucial ceramic of interest in the semiconductor, energy, and aerospace industries for its ability to perform in extreme conditions such as high temperatures and space travel. While the hexagonal 4H polytype is commonly used due to its wider band gap, beta SiC offers lower fabrication costs along with enhanced electron mobility and dielectric constant used for creating products such as sensors or transistors. By incorporating three distinct SiC morphologies—micron-sized spheres, whisker-like rods, and nano-sized spheres—into a polymer or ceramic matrix (PMMA or alpha alumina), this work examines how variations in geometric configurations (size, shape, dispersion, volume fraction) can affect the electrical and dielectric responses of the different matrix materials. Fabrication is carried out using pressure-only hydraulic pressing of the individual powders, compression molding of the PMMA matrix composites, and spark plasma sintering of the alumina matrix composites. Impedance & dielectric spectroscopy provides in-situ and post-fabrication electromagnetic characterization, which substantially broadens the understanding of observed trends in inductive and capacitive behavior that affect device performance. Confirmed during this study is the mechanism behind SiC microwave absorption, a useful property for electromagnetic interference (EMI) shielding that can be used to protect electronics such as that of satellites systems. This was tested using a vector network analyzer via a designed and assembled waveguide setup for measuring 2-port scattering parameters of samples. Statistical methodologies for broadband frequency measurements are demonstrated, useful for determining repeatability and reproducibility in quality analysis. Quality assessment was validated by working with interdigitated electrodes (IDEs) printed on composite FR4 substrates that serve as sensors for environmental conditions. The application of finite element analysis with COMSOL Multiphysics facilitated a more generalizable assessment approach, allowing for optimization of composite designs before large-scale production. This research thus provides a comprehensive framework linking processing methods and microstructure to electronic performance and quality assurance in SiC-based devices.
]]>
Date: Monday, May 18, 2026
Time: 1PM to 3PM Eastern Time (US)
Location: Coda 114 (1st floor conference room; just walk in, no special access needed)
Virtual Meeting: https://gatech.zoom.us/j/91061621484
Seongmin Lee
CS Ph.D. Candidate
School of Computational Science and Engineering
College of Computing
Georgia Institute of Technology
Committee:
Dr. Duen Horng (Polo) Chau - Advisor, Georgia Tech, School of Computational Science & Engineering
Dr. Alex Endert - Georgia Tech, School of Interactive Computing
Dr. Chao Zhang - Georgia Tech, School of Computational Science & Engineering
Dr. Judy Hoffman - University of California, Irvine, Donald Bren School of Information and Computer Sciences
Dr. Oliver Brdiczka - Adobe, Adobe Firefly
Abstract:
As modern AI systems, such as diffusion-based generative models or large language models (LLMs), continue to grow in scale, complexity, and societal impact, understanding and mitigating their risks has become increasingly urgent yet challenging due to their black-box nature.
My thesis addresses this critical challenge by developing novel visualizations and algorithms that help people understand the reasons and mechanisms behind AI behaviors, and take actionable steps to mitigate risks. Our work is organized into three complementary
thrusts:
(1) Attribute risks. We begin with investigating how to uncover the underlying causes of AI risks. We present the first survey bridging LLM interpretation and safety. Building on insights from our survey that training data can offer intuitive explanations for LLM generations, we develop LLM Attributor, which visually reveals the training data sources behind LLM-generated text, offering a novel way to diagnose unsafe outputs.
(2) Explain failure. While interpretation algorithms reveal causes of AI risks, their impact depends on how effectively they are communicated. To fill this gap, we introduce interactive visualizations that explain complex model mechanisms to broad audiences. Diffusion Explainer helps non-experts understand modern generative AI, outperforming traditional tools in 56-participant user studies. Extending visualization to non-generative models, VisCUIT empowers experts to explore the mechanisms behind failures of classifiers.
(3) Guide mitigation. To reduce risks, we introduce CRAYON, simple yet powerful algorithms that help classifiers overcome reliance on irrelevant features using yes-no annotations; experiments with large-scale human evaluations with 5,893 participants show its superiority over 12 methods across three datasets — even those requiring complex annotations. Extending to modern LLMs, we develop SHINE algorithm to determine whether hallucinations stem from limited model knowledge or flawed generation strategies. SHINE effectively differentiates faithful text and two types of hallucinations across three LLMs, and outperforms seven hallucination detection methods across four datasets and four LLMs.
My PhD research develops practical, innovative, human-centered solutions for research problems grounded in real-world needs, from advancing AI education to improving LLM safety, leveraging close partnership with leading companies like Google, Adobe, Cisco, JPMorgan Chase, ADP, and Avast. My work has made significant impacts across academia, industry, and society: Diffusion Explainer and its followup work Transformer Explainer have reached over 638k users in 210+ countries and have been integrated into university AI courses (e.g., MIT, Columbia). My research has been recognized with honors including the Korean Honor Scholarship, NCWIT AiC Collegiate Award Finalist, and IEEE VIS Best Poster Award.
]]>
Advisor: Dr. Dimitri Mavris
Milestone: PhD Thesis Proposal
Degree Program: Aerospace Engineering
Title: Multidisciplinary and Multifidelity Analysis and Design Optimization Framework for Twin Web Turbine Disk Technologies
Abstract: The continued push for higher thermal efficiency and reduced fuel consumption in modern gas turbine engines has led to increasingly aggressive operating conditions, including higher turbine inlet temperatures, elevated blade loading, and increased rotational speeds. While advances in materials, aerodynamics, and cooling technologies have enabled these trends, their full potential remains constrained by the structural capability of rotating components. Among these, the turbine disk is a critical limiting element, as it must sustain severe centrifugal loads and thermal gradients while interacting with complex cooling flows. This has motivated the development of advanced disk architectures such as the twin-web turbine disk (TWD), originally explored under the U.S. Air Force’s Composite Ring Reinforced Turbine (CRRT) program. However, despite its promise, the existing body of research on TWDs remains limited and fragmented across structural, thermal, and manufacturing domains. The design of TWDs is inherently multidisciplinary, requiring the coupled treatment of thermo-fluid-structural interactions alongside practical considerations such as manufacturability and cost. Current approaches often rely on partitioned, black-box workflows that require inter-solver data transfer, introducing additional sources of error and computational overhead. Furthermore, high-fidelity multiphysics simulations, particularly for cooling analysis, are computationally expensive, limiting their direct use in optimization. These challenges highlight the need for a unified framework that can consistently capture coupled physics while remaining computationally tractable. Motivated by these gaps, this work proposes the development of a unified, multidisciplinary, and multifidelity framework for the analysis and design optimization of TWDs. The proposed approach will systematically progress from a baseline partitioned formulation to a fully coupled monolithic solver, with the aim of improving numerical stability and predictive accuracy. To address computational cost, targeted surrogate models, particularly for cooling-related analyses, will be developed and incorporated to enable rapid yet reliable evaluations within the design loop. Sensitivity analysis and design space exploration will be carried out to identify the dominant design drivers, and the resulting framework will be embedded within a multidisciplinary optimization environment to generate Pareto-optimal designs. Through these efforts, this research aims to help bridge the gap between conceptual design and practical implementation of improved turbine disk technologies by combining high-fidelity multiphysics modeling with effective optimization strategies in a coherent formulation.
Date and time: 2026-05-15, 9:30 AM EDT
Location: CoVE
Committee:
Dr. Dimitri Mavris (advisor), School of Aerospace Engineering
Dr. Jechiel Jagoda, School of Aerospace Engineering
Dr. Kai James, School of Aerospace Engineering
Dr. Jonathan C. Gladin, School of Aerospace Engineering
Advisor: Dr. Adam Steinberg
Milestone: PhD Thesis Proposal
Degree Program: Aerospace Engineering
Title: Flame Structure, Operability, and Diagnostic Development in Lean Premixed Prevaporized Gas Turbine Combustors
Abstract: This proposal investigates flame structure, operability, and diagnostic development in lean premixed prevaporized (LPP) gas turbine combustors across operating regimes relevant to future low-emissions aircraft engines. LPP combustion offers a pathway to reduced NOx and non-volatile particulate matter emissions by burning globally lean. However, practical implementation is limited by competing operability challenges, including fuel-dependent mixing and vaporization effects, lean blowoff, and high-power failure modes such as flashback and upstream flame holding. This work addresses these challenges through three coordinated experimental campaigns using advanced optical diagnostics in both industrially relevant and laboratory-scale combustors. The first campaign examines how conventional Jet-A, hydroprocessed esters and fatty acids (HEFA) sustainable aviation fuel, and a 50/50 Jet-A–HEFA blend influence spray behavior, flame structure, and lean operability in a multi-element LPP combustor. Measurements using OH planar laser-induced fluorescence (PLIF), OH* chemiluminescence (CL), Mie scattering, and phase Doppler particle analysis are used to connect fuel physical properties to droplet penetration, vaporization, flame topology, and blowoff behavior. Preliminary results show that HEFA produces shorter liquid-fuel penetration depths than Jet-A and the blend, consistent with its lower density, viscosity, surface tension, and distillation characteristics. These results provide a basis for determining whether fuel-dependent spray and mixing differences significantly alter global lean blowoff behavior or primarily affect local flame structure and intermittency. The second campaign investigates pilot–main flame interactions in a practical multi-element LPP combustor by independently varying pilot and global equivalence ratios while acquiring measurements across multiple radial planes. OH/kerosene PLIF, Mie scattering, and high-speed OH* or CH* CL are used to identify how the central pilot modifies the surrounding bluff-body-stabilized main flames. This campaign is designed to determine whether the pilot primarily stabilizes the mains by supplying hot products and chemically active species to the main-flame recirculation zones, and whether global lean blowoff limits are more strongly controlled by main-flame conditions than by pilot equivalence ratio within the pilot’s own stability limits. The third campaign develops laser-induced phosphor thermometry (LIPT) for spatially and temporally resolved wall-temperature measurements under high-power takeoff conditions. Candidate phosphor coatings will be selected, calibrated, and evaluated for lifetime-based thermometry in a medium-pressure burner. LIPT measurements, combined with chemiluminescence imaging, will be used to identify localized wall-temperature signatures associated with abnormal flame–wall interactions and potential upstream flame holding. Although the medium-pressure burner cannot fully reproduce the thermal and optical environment of high-power testing, it provides a controlled platform for developing diagnostic capability that can later be extended to more severe gas-turbine operating conditions. Together, these campaigns aim to clarify how fuel properties, pilot–main coupling, and wall thermal response influence LPP combustor operability. The expected outcome is an improved physical understanding of flame stabilization and failure mechanisms in practical lean-burn combustors, along with diagnostic tools needed to characterize these processes across operating regimes.
Date and time: 2026-05-15, 2:00 PM
Location: Montgomery Knight Room 317
Committee:
Dr. Adam Steinberg (advisor), School of Aerospace Engineering
Dr. Ellen Mazumdar, School of Mechanical Engineering
Dr. Benjamin Emerson, School of Aerospace Engineering
Applied Physiology Thesis Defense
Henry Chionuma
School of Biological Sciences
Advisor:
Dr. Flavio Fenton
Georgia Institute of Technology
School of Physics
Open to the Community
The dispersion of action potential duration and intracellular calcium as a function of period of stimulation in explanted human hearts using optical mapping
May 14th, 2026
1:00pm
Howey Physics Building (837 State Street NW), Room C204
Committee Members:
Dr. Edward Balog, School of Biological Sciences; Georgia Institute of Technology
Dr. Elizabeth Cherry, School of Physics; Georgia Institute of Technology
Dr. Mindy Millard-Stafford, School of Biological Sciences; Georgia Institute of Technology
Dr. T. Richard Nichols, School of Biological Sciences [Professor Emeritus]; Georgia Institute of Technology
Background: Cardiovascular disease represents the leading cause of death globally, killing an estimated 17.9 million individuals each year. One of the most serious types of cardiovascular disease is cardiac arrhythmia, or ‘abnormal’ beating of the heart. Cardiac electrical wave disruption is the causative factor behind several of the most severe forms of cardiac arrhythmia, including fibrillation and tachycardia. Treatments for these arrhythmias are generally not optimal, owing in part to the fact that the mechanism of these arrhythmias are not well understood. Alternans, defined as periodic beat-to-beat oscillation in cardiac electrical activity and contraction strength, is as an important initiating event for tachyarrhythmias. Alternans can be detected electrocardiographically as T-wave alternans, a condition associated with a high mortality rates when not treated. At the cellular level, alternans can present as alternating long-short patterns of action potential duration (APD) and/or in alternating peak values of intracellular calcium ([Ca]i) concentration. Alternans in space dynamically increases dispersion of repolarization across the heart, which leads to large variations in refractory periods and conduction blocks which induce wave break and fibrillation.
Objectives: The proposed research will address the mechanisms of initiation of electrical arrhythmias driven by dynamically induced heterogeneities of refractoriness produced by alternans. Overall, the proposed project aims to increase the understanding of the mechanisms of human fibrillation, which could lead to the development of improved pharmacological or electrical interventions. Moreover, utilization of detailed optical-mapping data from human heart subjects has the potential to provide further insight into how experimental results in other mammalian hearts can be applied to understand human heart mechanisms. Similarly, insight from the proposed research can help to improve computational modeling of arrhythmias in human hearts through improved calibration and validation.
Specific aims: 1: Perform the first detailed spatiotemporal quantification of APD and [Ca]i dispersion in explanted human hearts across many physiological periods of stimulation. 2: Compare APD and [Ca]i dispersion between human and non-human hearts. 3: Quantify the effects of temperature on dispersion of APD and [Ca]i.
Research design and methods: Simultaneous optical mapping of voltage and calcium will be used. Human hearts will be obtained through an existing collaboration with the Emory University Hospital transplant program and non-human hearts will be obtained from other terminal studies. Hearts will be stimulated using established protocols where rates are increased to induce alternans followed by either initiation of fibrillation or conduction block. Computational models of human heart cells and tissue will be used for comparison with experiments.
]]>
Advisor: Prof. Josh Kacher
will propose a doctoral thesis entitled,
Hydrogen Reduction of Mixed Metal Oxides for Solid-State Alloy Formation of Stainless Steel
On
Thursday, May 14 at 9:30 a.m.
Love Room 295
And
Virtually via MS Teams
https://teams.microsoft.com/meet/215969704040016?p=93wAJQ4kuW2sm3JNv5
Committee
Prof. Josh Kacher – School of Materials Science and Engineering (advisor)
Prof. Naresh Thadhani– School of Materials Science and Engineering
Prof. Robert Speyer – School of Materials Science and Engineering
Prof. Preet Singh – School of Materials Science and Engineering
Prof. Chaitanya Deo – School of Mechanical Engineering
Abstract
Hydrogen reduction of metal oxides offers a promising pathway to carbon- and energy-efficient ferrous alloy and steel production, but the high cost of hydrogen often makes this route economically untenable. One path forward is a process-intensification approach involving extrusion-based net-shape fabrication of alloy steels, starting with oxide components, followed by their reduction and sintering, resulting in the metal/alloy product. In this work, we study the coreduction behavior of the binary Fe-18Cr alloy, relevant to the common 316L stainless steel system. A central challenge is the reduction of chromium oxide (Cr₂O₃), a highly stable oxide that often remains partially unreduced, thereby limiting alloy homogeneity. While the early reduction of iron oxide (Fe₂O₃) can promote chromia reduction by acting as a sink for newly formed chromium, it can also hinder complete reduction by driving extensive Fe sintering, which encapsulates residual Cr oxides and restricts gas transport, thereby limiting further reduction. Preliminary results show that chromia particle size and sample thickness strongly influence reduction behavior. This work aims to gain a deep understanding of the interplay among thermodynamics, gas transport, and microstructural evolution that governs coreduction, using advanced characterization techniques such as SEM, XRD, and TEM to probe phase and interfacial evolution. The findings will guide strategies to achieve full reduction and inform the design of oxide-dispersed steels through the controlled incorporation of stable nanoscale oxides, ultimately advancing a scalable, low-carbon pathway for steel production.
]]>
Committee:
Dr. Shaolan Li, ECE, Chair, Advisor
Dr. Levent Degertekin, ECE, Co-Advisor
Dr. Omer Inan, ECE
Dr. Visvesh Sathe, ECE
Dr. Jane Gu, ECE
Dr. Coskun Tekes, KSU
]]>Student Name: McKenna Clinch
Thesis Title: Ingestible Electroceutical Device for Satiety and Digestion Hormone Regulation
Thesis Advisor: Alex Abramson
Thesis Co-Advisor: N/A
Committee Members: Alex Abramson - Chemical and Biomolecular Engineering; Hang Lu - Chemical and Biomolecular Engineering; Michael Filler - Chemical and Biomolecular Engineering; Matthew Flavin - Electrical and Computer Engineering; Shanthi Srinivasan - Department of Medicine (Emory)
Date: 5/19/2026
Time: 1:00 PM
Location: MoSE 3201A
]]>School of Psychology – Ph.D. Dissertation Proposal Meeting
Date: Wednesday, May 13th, 2026
Time: 12:00 - 2:00PM
Location: JS Coon 148 or online (https://teams.microsoft.com/meet/215526552621487?p=mdNo1jdq8KNDRi5KvS)
Dissertation Committee Chair/Advisor:
Bruce Walker, Ph.D. (Georgia Tech)
Dissertation Committee Members:
Richard Catrambone, Ph.D. (Georgia Tech)
Woon Ju Park, Ph.D. (Georgia Tech)
Ben Satterfield, Ph.D. (Georgia Tech)
Barbara Chaparro, Ph.D. (Embry-Riddle Aeronautical University)
Title: From Acceptance to Mastery: An Evaluation of Technology Mastery and the Validation of the General Technology Mastery Scale
Abstract: The goal of this proposed dissertation is to empirically investigate the concept of technology mastery, operationally defined here as an individual’s knowledge, attitudes, and ability regarding the use of a technology. To begin, I conducted a series of interviews with a mix of community members and college undergraduates to better define the concept of “technology mastery,” determine what behaviors and traits are associated with the term, and finally discuss any barriers individuals face when attempting to gain skill with technology. The results of this study were analyzed through thematic analysis and produced roughly 40 themes, which were then used to create the General Technology Mastery Scale (GTMS) through a series of factor analysis techniques. I then sought to validate the GTMS by comparing it to another available scale of technology mastery, the Continuum of Assistive Technology Mastery (CATM). Additionally, I collected data on hypothesized covariates based on the literature (such as problem-solving, technology self-efficacy, general self-efficacy, technology acceptance, system usability, locus of control, motivation, and creativity). For the proposed Study 4, I plan to collect data at four timepoints on a sample of master's level speech language pathology students at the University of Georgia who are enrolled in CMSD 6650 (a course on Augmentative and Alternative Communication Devices). This combined body of work introduces the most comprehensive explanation of technology mastery and its correlates to date. Furthermore, it proposes a quick (if imperfect) way to measure technology mastery when collecting data on objective performance is not feasible
]]>Committee:
Dr. Yorai Wardi, ECE, Chair, Advisor
Dr. Fumin Zhang, ECE
Dr. Matthieu Bloch, ECE
Dr. Maegan Tucker, ECE
Dr. Haomin Zhou, Math
]]>Title: Tensor-based Predictive Modeling and Control for High-dimensional Data
Date: May 15, 2026 (Friday)
Time: 10:00 am - 12:00 pm EST
Zoom link:
(Meeting ID: 971 7870 0611; Passcode: 387776)
Zihan Zhang
Ph.D. Candidate in Industrial Engineering
H. Milton Stewart School of Industrial and Systems Engineering
Georgia Institute of Technology
Thesis Committee:
Abstract:
Advances in sensing technologies have dramatically increased the volume and complexity of high-dimensional data, such as high-resolution images and videos, that challenge the foundations of traditional control methodologies. Conventional approaches, rooted in low-dimensional signal processing, often struggle to capture the complex spatio-temporal dependencies inherent in such data. Naive vectorization techniques destroy essential structural information, while many learning-based methods require large datasets and often lack interpretability.
This dissertation develops tensor-based control frameworks that preserve the underlying spatial and temporal structure of high-dimensional data. Chapter 2 addresses incomplete sensing in automatic process control by introducing methods for response imputation and control under missing-data conditions. Chapter 3 presents a system modeling framework that captures localized correlations in system responses and the spatial effects of control actions, followed by a dynamic controller design that optimizes controller placement to improve performance. Chapter 4 incorporates diffusion models to capture nonlinear spatio-temporal correlations and enable uncertainty quantification.
Together, these contributions advance a data-efficient and interpretable framework for controlling intelligent systems that operate with high-dimensional, multimodal sensory data.
Committee:
Dr. Sathe, Advisor
Dr. Romberg, Chair
Dr. Raychowdhury
]]>