I2S Masters/ Doctoral Theses


All students and faculty are welcome to attend the final defense of I2S graduate students completing their M.S. or Ph.D. degrees. Defense notices for M.S./Ph.D. presentations for this year and several previous years are listed below in reverse chronological order.

Students who are nearing the completion of their M.S./Ph.D. research should schedule their final defenses through the EECS graduate office at least THREE WEEKS PRIOR to their presentation date so that there is time to complete the degree requirements check, and post the presentation announcement online.

Upcoming Defense Notices

Arya Hadizadeh Moghaddam

Learning Personalized and Robust Patient Representations across Complex Structures in Electronic Health Records

When & Where:


Eaton Hall, Room 2001B

Degree Type:

PhD Dissertation Defense

Committee Members:

Zijun Yao, Chair
Fengjun Li
Bo Luo
Han Wang
Xinmai Yang

Abstract

Electronic health records (EHRs) contain medical codes and clinical notes collected over multiple visits. This dissertation studies how to learn patient representations from these records through three related aspects problems:

The first aspect concerns relationships within a patient’s history. TPGT models changing dependencies among medical codes within and across visits, and ARCI separates temporal patterns associated with different medical intents. They account for interacting clinical events and concurrent treatment goals.

When patient histories are limited, these patterns are harder to learn. The second aspect is addressed by MetaDrug, which adapts to patients with sparse histories using their own records and those of similar patients, while filtering uncertain samples.

The third aspect concerns incorporating clinical notes and large language models into patient representations. RePrompT uses cohort-trained representations and recurrent soft prompts to preserve longitudinal context. RASPER trains a note summarizer through downstream prediction feedback. CERE reasons over notes conditioned on structured codes to represent the additional evidence they provide. These studies connect structured histories with clinical text to support prediction.

Experiments on the KUMC acute kidney injury cohort, eICU, MIMIC-III, and MIMIC-IV show improvements over the respective baselines in clinical prediction and medication recommendation.


Shelby Jones

Measuring LLM Susceptibility to Stylistic Perturbations in Social Media Text

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

MS Thesis Defense

Committee Members:

Jennifer Lohoefener, Chair
Perry Alexander
John Symons


Abstract

Large Language Models (LLMs) are becoming increasingly integrated into important domains such as social media and politics. This makes it essential that we understand their susceptibility to superficial text variations. In this thesis, I investigated whether small stylistic perturbations in social media posts influence how an LLM perceives a post's political leanings. To do this, I systematically manipulated punctuation and capitalization across both political and apolitical social media posts. I then used VADER sentiment analysis to measure the changes in semantic intensity to quantify the text manipulation. Finally, I prompted each LLM to provide its perceived political leanings of each given post. I used linear regression to assess the VADER score correlation with the LLMs' political alignment ratings. While there was no statistically significant relation between perceived political extremity and VADER scores, category-level analysis revealed that the stylistic perturbations cause significant inconsistencies in model outputs. Furthermore, in posts with neutral semantic content, the stylistic perturbations introduce a right-leaning bias. These findings add to the growing body of literature on LLM instability under prompt perturbation and highlight risks associated with automated content analysis. 


Connor Sullivan

Memory QoS Mechanisms for Predictable and Performant Real-Time Systems

When & Where:


Eaton Hall, Room 2001B

Degree Type:

PhD Comprehensive Defense

Committee Members:

Heechul Yun, Chair
Zitao Chen
Tamzidul Hoque
Prasad Kulkarni
Kenneth Fischer

Abstract

Commercial-off-the-shelf (COTS) multicore processors are increasingly used in safety-critical real-time systems due to their higher performance and improved size, weight, and power (SWaP) characteristics. However, modern platforms are designed for average-case performance, which works in direct opposition to the worst-case timing guarantees required by safety-critical systems.

In particular, unregulated sharing of the memory hierarchy across cores can cause severe and unpredictable interference. When multiple cores concurrently access shared memory resources, their requests contend with one another, degrading application performance and undermining timing guarantees. Hardware quality-of-service (QoS) mechanisms aim to manage this contention. Partitioning these resources in space alone does not guarantee temporal isolation. Bandwidth regulation can protect latency-critical tasks from such interference, but it trades system throughput for isolation. Current solutions throttle co-running tasks far more than necessary, offering system designers a poor isolation-throughput tradeoff.

Addressing this challenge, this thesis focuses on QoS mechanisms that improve the balance between throughput and isolation. Shared memory resources such as the last-level cache (LLC) and main memory (DRAM) have parallel, banked structures, yet existing regulation treats them as monolithic. This thesis shows that contention in both arises at the bank level and proposes per-bank bandwidth regulation for the LLC and DRAM, which preserves isolation for latency-critical tasks while recovering throughput for best-effort tasks. Finally, the thesis aims to coordinate regulation across the levels of the memory hierarchy, adapting it dynamically at runtime. It will also explore matching the control mechanism to the root cause of contention, for example by controlling the number of outstanding requests, as an alternative or cooperative means to bandwidth regulation. Together, these efforts work toward a better balance between system throughput and isolation guarantees in multicore real-time systems.


Fairuz Shadmani Shishir

Efficient and Privacy-Preserving Biomedical AI: From Precision Therapeutics Design to Private Clinical Prediction

When & Where:


Eaton Hall, Room 2001B

Degree Type:

PhD Dissertation Defense

Committee Members:

Sumaiya Shomaji, Chair
Tamzidul Hoque
Cuncong Zhong
Bishnu Sarker
Michael Hageman

Abstract

Accurate protein characterization, precision therapeutics design, and privacy-preserving clinical modeling are important challenges in biomedical artificial intelligence (AI). Addressing these challenges requires computational frameworks that are not only accurate and scalable but also capable of supporting reliable decision-making. This dissertation develops efficient, scalable, and reliable computational frameworks across three areas: Computational characterization of proteins, precision therapeutics design, and privacy-aware clinical representation learning. The first part of this dissertation focuses on efficient protein characterization at both the sequence and residue levels. GPCR-SLM, a knowledge-distilled small protein language model, classifies G-protein-coupled receptors (GPCRs), a family targeted by approximately 35% of FDA-approved drugs. Across 86 families, GPCR-SLM achieves 99% accuracy, outperforming BLAST (86.4%) and HMMER (91%), with a 33.5-fold speedup over large protein language models. Extending characterization to the residue level, a protein-metal-ion binding site prediction framework evaluates five protein language models and incorporates positional encoding to capture long-range dependencies. Under 10-fold cross-validation, it achieves a Matthews correlation coefficient of 0.89, with precision, recall, and F1 scores exceeding 95% across six metal ions.

Building on these protein characterization methods, the second part advances precision therapeutics design through uncertainty quantification and generative learning for therapeutic antibodies. NbBayesLM combines protein language model embeddings and physicochemical descriptors in a Bayesian neural network to predict nanobody thermostability. Evaluated on 10,630 VHH sequences, it achieves a mean absolute error of 1.89C and an R2 of 0.67 while estimating epistemic and aleatoric uncertainty to support candidate prioritization. CIgFlow employs classifier-free guidance to generate de novo, antigen-conditioned antibody sequences without fixed scaffolds or paired structural data. Generated sequences achieve a median OASis humanness score of 0.90 and mean predicted local distance difference test (pLDDT) score of 85.0, demonstrating promising novelty and diversity. The third part addresses privacy-aware clinical modeling using electrocardiograms (ECGs), which encode diagnostic information and sensitive demographic attributes. A variational autoencoder with dual discriminators suppresses demographic information while preserving clinically relevant representations.

Collectively, this dissertation advances efficient protein representation learning, computational antibody design, and privacy-preserving clinical AI, contributing methodological foundations for trustworthy biomedical applications.


Past Defense Notices

Dates

Rokunuz Jahan Rudro

Using Machine Learning to Classify Driver Behavior from Psychological Features: An Exploratory Study

When & Where:


Eaton Hall, Room 1A

Degree Type:

MS Thesis Defense

Committee Members:

Sumaiya Shomaji, Chair
David Johnson
Zijun Yao
Alexandra Kondyli

Abstract

Driver inattention and human error are the primary causes of traffic crashes. However, little is known about the relationship between driver aggressiveness and safety. Although several studies that group drivers into different classes based on their driving performance have been conducted, little has been done to explore how behavioral traits are linked to driver behavior. The study aims to link different driver profiles, assessed through psychological evaluations, with their likelihood of engaging in risky driving behaviors, as measured in a driving simulation experiment. By incorporating psychological factors into machine learning algorithms, our models were able to successfully relate self-reported decision-making and personality characteristics with actual driving actions. Our results hold promise toward refining existing models of driver behavior by understanding the psychological and behavioral characteristics that influence the risk of crashes.


Md Mashfiq Rizvee

Energy Optimization in Multitask Neural Networks through Layer Sharing

When & Where:


Eaton Hall, Room 2001B

Degree Type:

MS Thesis Defense

Committee Members:

Sumaiya Shomaji, Chair
Tamzidul Hoque
Han Wang


Abstract

Artificial Intelligence (AI) is being widely used in diverse domains such as industrial automation, traffic control, precision agriculture, and smart cities for major heavy lifting in terms of data analysis and decision making. However, the AI life- cycle is a major source of greenhouse gas (GHG) emission leading to devastating environmental impact. This is due to expensive neural architecture searches, training of countless number of models per day across the world, in-field AI processing of data in billions of edge devices, and advanced security measures across the AI life cycle. Modern applications often involve multitasking, which involves performing a variety of analyzes on the same dataset. These tasks are usually executed on resource-limited edge devices, necessitating AI models that exhibit efficiency across various measures such as power consumption, frame rate, and model size. To address these challenges, we introduce a novel neural network architecture model that incorporates a layer sharing principle to optimize the power usage. We propose a novel neural architecture, Layer Shared Neural Networks that merges multiple similar AI/NN tasks together (with shared layers) towards creating a single AI/NN model with reduced energy requirements and carbon footprint. The experimental findings reveal competitive accuracy and reduced power consumption. The layer shared model significantly reduces power consumption by 50% during training and 59.10% during inference causing as much as an 84.64% and 87.10% decrease in CO2 emissions respectively. 

  


Fairuz Shadmani Shishir

Parameter-Efficient Computational Drug Discovery using Deep Learning

When & Where:


Eaton Hall, Room 2001B

Degree Type:

MS Thesis Defense

Committee Members:

Sumaiya Shomaji, Chair
Tamzidul Hoque
Hongyang Sun


Abstract

The accurate prediction of small molecule binding affinity and toxicity remains a central challenge in drug discovery, with significant implications for reducing development costs, improving candidate prioritization, and enhancing safety profiles. Traditional computational approaches, such as molecular docking and quantitative structure-activity relationship (QSAR) models, often rely on handcrafted features and require extensive domain knowledge, which can limit scalability and generalization to novel chemical scaffolds. Recent advances in language models (LMs), particularly those adapted to chemical representations such as SMILES (Simplified Molecular Input Line Entry System), have opened new ways for learning data-driven molecular representations that capture complex structural and functional properties. However, achieving both high binding affinity and low toxicity through a resource-efficient computational pipeline is inherently difficult due to the multi-objective nature of the task. This study presents a novel dual-paradigm approach to critical challenges in drug discovery: predicting small molecules with high binding affinity and low cardiotoxicity profiles. For binding affinity prediction, we implement a specialized graph neural network (GNN) architecture that operates directly on molecular structures represented as graphs, where atoms serve as nodes and bonds as edges. This topology-aware approach enables the model to capture complex spatial arrangements and electronic interactions critical for protein-ligand binding. For toxicity prediction, we leverage chemical language models (CLMs) fine-tuned with Low-Rank Adaptation (LoRA), allowing efficient adaptation of large pre-trained models to specialized toxicological endpoints while maintaining the generalized chemical knowledge embedded in the base model. Our hybrid methodology demonstrates significant improvements over existing computational approaches, with the GNN component achieving an average area under the ROC curve (AUROC) of 0.92 on three protein targets and the LoRA-adapted CLM reaching (AUROC) of 0.90 with 60% reduction in parameter usage in predicting cardiotoxicity. This work establishes a powerful computational framework that accelerates drug discovery by enabling both higher binding affinity and low toxicity compounds with optimized efficacy and safety profiles. 


Soma Pal

Truths about compiler optimization for state-of-the-art (SOTA) C/C++ compilers

When & Where:


Eaton Hall, Room 2001B

Degree Type:

PhD Comprehensive Defense

Committee Members:

Prasad Kulkami, Chair
Esam El-Araby
Drew Davidson
Tamzidul Hoque
Jian Yunfeng

Abstract

Compiler optimizations are critical for performance and have been extensively studied, especially for C/C++ language compilers. Our overall goal in this thesis is to investigate and compare the properties and behavior of optimization passes across multiple contemporary, state-of-the-art (SOTA)  C/C++ compilers to understand if they adopt similar optimization implementation and orchestration strategies. Given the maturity of pre-existing knowledge in the field, it seems conceivable that different compiler teams will adopt consistent optimization passes, pipeline and application techniques. However, our preliminary results indicate that such expectation may be misguided. If so, then we will attempt to understand the differences, and study and quantify their impact on the performance of generated code.

In our first work, we study and compare the behavior of profile-guided optimizations (PGO) in two popular SOTA C/C++ compilers, GCC and Clang. This study reveals many interesting, and several counter-intuitive, properties about PGOs in C/C++ compilers. The behavior and benefits of PGOs also vary significantly across our selected compilers. We present our observations, along with plans to further explore these inconsistencies in this report. Likewise, we have also measured noticeable differences in the performance delivered by optimizations across our compilers. We propose to explore and understand these differences in this work. We present further details regarding our proposed directions and planned experiments in this report. We hope that this work will show and suggest opportunities for compilers to learn from each other and motivate researchers to find mechanisms to combine the benefits of multiple compilers to deliver higher overall program performance.


Nyamtulla Shaik

AI Vision to Care: A QuadView of Deep Learning for Detecting Harmful Stimming in Autism

When & Where:


Eaton Hall, Room 2001B

Degree Type:

MS Thesis Defense

Committee Members:

Sumaiya Shomaji, Chair
Bo Luo
Dongjie Wang


Abstract

Stimming refers to repetitive actions or behaviors used to regulate sensory input or express feelings. Children with developmental disorders like autism (ASD) frequently perform stimming. This includes arm flapping, head banging, finger flicking, spinning, etc. This is exhibited by 80-90% of children with Autism, which is seen in 1 among 36 children in the US. Head banging is one of these self-stimulatory habits that can be harmful. If these behaviors are automatically identified and notified using live video monitoring, parents and other caregivers can better watch over and assist children with ASD.

Classifying these actions is important to recognize harmful stimming, so this study focuses on developing a deep learning-based approach for stimming action recognition. We implemented and evaluated four models leveraging three deep learning architectures based on Convolutional Neural Networks (CNNs), Autoencoders, and Vision Transformers. For the first time in this area, we use skeletal joints extracted from video sequences. Previous works relied solely on raw RGB videos, vulnerable to lighting and environmental changes. This research explores Deep Learning based skeletal action recognition and data processing techniques for a small unstructured dataset that consists of 89 home recorded videos collected from publicly available sources like YouTube. Our robust data cleaning and pre-processing techniques helped the integration of skeletal data in stimming action recognition, which performed better than state-of-the-art with a classification accuracy of up to 87%.

In addition to using traditional deep learning models like CNNs for action recognition, this study is among the first to apply data-hungry models like Vision Transformers (ViTs) and Autoencoders for stimming action recognition on the dataset. The results prove that using skeletal data reduces the processing time and significantly improves action recognition, promising a real-time approach for video monitoring applications. This research advances the development of automated systems that can assist caregivers in more efficiently tracking stimming activities.
 


Alexander Rodolfo Lara

Creating a Faradaic Efficiency Graph Dataset Using Machine Learning

When & Where:


Eaton Hall, Room 2001B

Degree Type:

MS Project Defense

Committee Members:

Zijun Yao, Chair
Sumaiya Shomaji
Kevin Leonard


Abstract

Just as the internet-of-things leverages machine learning over a vast amount of data produced by an innumerable number of sensors, the Internet of Catalysis program uses similar strategies with catalysis research. One application of the Internet of Catalysis strategy is treating research papers as datapoints, rich with text, figures, and tables. Prior research within the program focused on machine learning models applied strictly over text.This project is the first step of the program in creating a machine learning model from the images of catalysis research papers. Specifically, this project creates a dataset of faradaic efficiency graphs using transfer learning from pretrained models. The project utilizes FasterRCNN_ResNet50_FPN, LayoutLMv3SequenceClassification, and computer vision techniques to recognize figures, extract all graphs, then classify the faradaic efficiency graphs.

Downstream of this project, researchers will create a graph reading model to integrate with large language models. This could potentially lead to a multimodal model capable of fully learning from images, tables, and texts of catalysis research papers. Such a model could then guide experimentation on reaction conditions, catalysts, and production.


Amin Shojaei

Scalable and Cooperative Multi-Agent Reinforcement Learning for Networked Cyber-Physical Systems: Applications in Smart Grids

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

PhD Dissertation Defense

Committee Members:

Morteza Hashemi, Chair
Alex Bardas
Prasad Kulkarni
Taejoon Kim
Shawn Keshmiri

Abstract

Significant advances in information and networking technologies have transformed Cyber-Physical Systems (CPS) into networked cyber-physical systems (NCPS). A noteworthy example of such systems is smart grid networks, which include distributed energy resources (DERs), renewable generation, and the widespread adoption of Electric Vehicles (EVs). Such complex NCPS require intelligent and autonomous control solutions. For example, the increasing number of EVs introduces significant sources of demand and user behavior uncertainty that can jeopardize grid stability during peak hours. Traditional model-based demand-supply controls fail to accurately model and capture the complex nature of smart grid systems in the presence of different uncertainties and as the system size grows. To address these challenges, data-driven approaches have emerged as an effective solution for informed decision-making, predictive modeling, and adaptive control to enhance the resiliency of NCPS in uncertain environments.

As a powerful data-driven approach, Multi-Agent Reinforcement Learning (MARL) enables agents to learn and adapt in dynamic and uncertain environments. However, MARL techniques introduce complexities related to communication, coordination, and synchronization among agents. In this PhD research, we investigate autonomous control for smart grid decision networks using MARL. First, we examine the issue of imperfect state information, which frequently arises due to the inherent uncertainties and limitations in observing the system state.

Second, we focus on the cooperative behavior of agents in distributed MARL frameworks, particularly under the central training with decentralized execution (CTDE) paradigm. We provide theoretical results and variance analysis for stochastic and deterministic cooperative MARL algorithms, including Multi-Agent Deep Deterministic Policy Gradient (MADDPG), Multi-Agent Proximal Policy Optimization (MAPPO), and Dueling MAPPO. These analyses highlight how coordinated learning can improve system-wide decision-making in uncertain and dynamic environments like EV networks.

Third, we address the scalability challenge in large-scale NCPS by introducing a hierarchical MARL framework based on a cluster-based architecture. This framework organizes agents into coordinated subgroups, improving scalability while preserving local coordination. We conduct a detailed variance analysis of this approach to demonstrate its effectiveness in reducing communication overhead and learning complexity. This analysis establishes a theoretical foundation for scalable and efficient control in large-scale smart grid applications.


Asrith Gudivada

Custom CNN for Object State Classification in Robotic Cooking

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

MS Project Defense

Committee Members:

David Johnson, Chair
Prasad Kulkarni
Dongjie Wang


Abstract

This project presents the development of a custom Convolutional Neural Network (CNN) designed to classify object states—such as sliced, diced, or peeled—in robotic cooking environments. Recognizing fine-grained object states is critical for context-aware manipulation yet remains a challenging task due to the visual similarity between states and the limited availability of cooking-specific datasets. To address these challenges, we built a lightweight, non-pretrained CNN trained on a curated dataset of 11 object states. Starting with a baseline architecture, we progressively enhanced the model using data augmentation, optimized dropout, batch normalization, Inception modules, and residual connections. These improvements led to a performance increase from ~45% to ~52% test accuracy. The final model demonstrates improved generalization and training stability, showcasing the effectiveness of combining classical and advanced deep learning techniques. This work contributes toward real-time state recognition for autonomous robotic cooking systems, with implications for assistive technologies in domestic and elder care settings.


 


Hara Madhav Talasila

Radiometric Calibration of Radar Depth Sounder Data Products

When & Where:


Nichols Hall, Room 317 (Richard K. Moore Conference Room)

Degree Type:

PhD Dissertation Defense

Committee Members:

Carl Leuschen, Chair
Patrick McCormick
James Stiles
Jilu Li
Leah Stearns

Abstract

Although the Center for Remote Sensing of Ice Sheets (CReSIS) performs several radar calibration steps to produce Operation IceBridge (OIB) radar depth sounder data products, these datasets are not radiometrically calibrated and the swath array processing uses ideal (rather than measured [calibrated]) steering vectors. Any errors in the steering vectors, which describe the response of the radar as a function of arrival angle, will lead to errors in positioning and backscatter that subsequently affect estimates of basal conditions, ice thickness, and radar attenuation. Scientific applications that estimate physical characteristics of surface and subsurface targets from the backscatter are limited with the current data because it is not absolutely calibrated. Moreover, changes in instrument hardware and processing methods for OIB over the last decade affect the quality of inter-seasonal comparisons. Recent methods which interpret basal conditions and calculate radar attenuation using CReSIS OIB 2D radar depth sounder echograms are forced to use relative scattering power, rather than absolute methods.

As an active target calibration is not possible for past field seasons, a method that uses natural targets will be developed. Unsaturated natural target returns from smooth sea-ice leads or lakes are imaged in many datasets and have known scattering responses. The proposed method forms a system of linear equations with the recorded scattering signatures from these known targets, scattering signatures from crossing flight paths, and the radiometric correction terms. A least squares solution to optimize the radiometric correction terms is calculated, which minimizes the error function representing the mismatch in expected and measured scattering. The new correction terms will be used to correct the remaining mission data. The radar depth sounder data from all OIB campaigns can be reprocessed to produce absolutely calibrated echograms for the Arctic and Antarctic. A software simulator will be developed to study calibration errors and verify the calibration software. The software for processing natural targets and crossovers will be made available in CReSIS’s open-source polar radar software toolbox. The OIB data will be reprocessed with new calibration terms, providing to the data user community a complete set of radiometrically calibrated radar echograms for the CReSIS OIB radar depth sounder for the first time.


Christopher Ord

A Hardware-Agnostic Simultaneous Transmit And Receive (STAR) Architecture for the Transmission of Non-Repeating FMCW Waveforms

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

MS Thesis Defense

Committee Members:

Rachel Jarvis, Chair
Shannon Blunt
Patrick McCormick


Abstract

With the increasing congestion of the usable RF spectrum, it is increasingly necessary for communication and radar systems to share the same frequencies without disturbing one another. To accomplish this, research has focused on designing a class of non-repeating radar waveforms that appear as noise at the receiver of uncooperative systems, but the peak power from high-power pulsed systems can still overwhelm nearby in-band systems. Therefore, to minimize peak power while maximizing the total energy on target, radar systems must transition to operating at a 100% duty cycle, which inherently requires Simultaneous Transmit and Receive (STAR) operation.

One inherent difficulty when operating monostatic STAR systems is the direct path coupling interference that can saturate a number of components in the radar’s receive chain, which makes digital processing methods that remove this interference ineffective. This thesis proposes a method to reduce the self-interference between the radar’s transmitter in receiver prior to the receiver’s sensitive components to increase the power that the radar can transmit at. By using a combination of tests that manipulate the timing, phase, and magnitude of a secondary waveform that is injected into the radar just before the receiver, upwards of 35.0 dB of self-interference cancellation is achieved for radar waveforms with bandwidths of up to 100 MHz at both S-band and X-band in both simulation and open-air testing.