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

Neel Patel

Near-Memory Acceleration of Compressed Far Memory

When & Where:


Nichols Hall, Room 250 (Gemini Room)

Degree Type:

MS Thesis Defense

Committee Members:

Mohammad Allan, Chair
David Johnson
Prasad Kulkarni


Abstract

DRAM constitutes over 50% of server cost and 75% of the embodied carbon footprint of a server. To mitigate DRAM cost, far memory architectures have emerged. They can be separated into two broad categories: software-defined far memory (SFM) and disaggregated far memory (DFM). In this work, we compare the cost of SFM and DFM in terms of their required capital investment, operational expense, and carbon footprint. We show that, for applications whose data sets are compressible and have predictable memory access patterns, it takes several years for a DFM to break even with an equivalent capacity SFM in terms of cost and sustainability. We then introduce XFM, a near-memory accelerated SFM architecture, which exploits the coldness of data during SFM-initiated swap ins and outs. XFM leverages refresh cycles to seamlessly switch the access control of DRAM between the CPU and near-memory accelerator. XFM parallelizes near-memory accelerator accesses with row refreshes and removes the memory interference caused by SFM swap ins and outs.

We modify an open source far memory implementation to implement a full-stack, user-level XFM. Our experimental results use a combination of an FPGA implementation, simulation, and analytical modeling to show that XFM eliminates memory bandwidth utilization when performing compression and decompression operations with SFMs of capacities up to 1TB. The memory and cache utilization reductions translate to 5∼27% improvement in the combined performance of co-running applications.


Durga Venkata Suraj Tedla

Block chain based inter organization file sharing system

When & Where:


Eaton Hall, Room 2001B

Degree Type:

MS Project Defense

Committee Members:

David Johnson, Chair
Drew Davidson
Sankha Guria


Abstract

A coalition of companies collaborates collectively and shares information to improve their operations together. Distributed trust and transparency cannot be obtained with centralized file-sharing platforms. File sharing may be done transparently and securely with blockchain technology. This project suggests an inter-organizational secure file-sharing system based on blockchain technology. The group can use it to securely share files in a distributed manner. The creation of smart contracts and the configuration of blockchain networks are carried out by Hyperledger Fabric, an enterprise blockchain platform. Distributed file storage is accomplished through the usage of the Inter Planetary File System (IPFS).

The workflow for file-sharing and identity management procedures is provided in the paper. Using blockchain technology, the recommended approach enables a group of businesses to share files with availability, integrity, and confidentiality. The suggested method uses blockchain to enable safe file exchange amongst a group of enterprises. It offers shared file availability, confidentiality, and integrity. It guarantees complete file encryption. The blockchain provides tamper-resistant storage for the shared file's content ID. On the distributed storage and blockchain ledger, respectively, the encrypted file and file metadata are stored.


Dang Qua Nguyen

Hybrid Precoding Optimization and Private Federated Learning for Future Wireless Systems

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

PhD Comprehensive Defense

Committee Members:

Taejoon Kim, Chair
Morteza Hashemi
Erik Perrins
Zijun Yao
KC Kong

Abstract

This PhD research addresses two challenges in future wireless systems: hybrid precoder design for sub-Terahertz (sub-THz) massive multiple-input multiple-output (MIMO) communications and private federated learning (FL) over wireless channels. The first part of the research introduces a novel hybrid precoding framework that combines true-time delay (TTD) and phase shifters (PS) precoders to counteract the beam squint effect - a significant challenge in sub-THz massive MIMO systems that leads to considerable loss in array gain.

Our research presents a novel joint optimization framework for the TTD and PS precoder design, incorporating realistic time delay constraints for each TTD device. We first derive a lower bound on the achievable rate of the system and show that, in the asymptotic regime, the optimal analog precoder that fully compensates for the beam squint is equivalent to the one that maximizes this lower bound. Unlike previous methods, our framework does not rely on the unbounded time delay assumption and optimizes the TTD and PS values jointly to cope with the practical limitations. Furthermore, we determine the minimum number of TTD devices needed to reach a target array gain using our proposed approach.

Simulations validate that the proposed approach demonstrates performance enhancement, ensures array gain, and achieves computational efficiency. In the second part, the research devises a differentially private FL algorithm that employs time-varying noise perturbation and optimizes transmit power to counteract privacy risks, particularly those stemming from engineering-inversion attacks. This method harnesses inherent wireless channel noise to strike a balance between privacy protection and learning utility. By strategically designing noise perturbation and power control, our approach not only safeguards user privacy but also upholds the quality of the learned FL model. Additionally, the number of FL iterations is optimized by minimizing the upper bound on the learning error. We conduct simulations to showcase the effectiveness of our approach in terms of DP guarantee and learning utility.


Sai Narendra Koganti

Object Detection

When & Where:


Nichols Hall, Room 250 (Gemini Room)

Degree Type:

MS Project Defense

Committee Members:

Sumaiya Shomaji, Chair
David Johnson
Prasad Kulkarni


Abstract

This project offers a hands-on investigation of object identification utilizing the YOLO method, Python, and OpenCV. It begins by explaining the YOLO architecture, focusing on the single-stage detection process for bounding box prediction and class probability calculation. The setup phase includes library installation and model configuration, resulting in a smooth implementation procedure. Using OpenCV, the project includes preparatory processes required for object detection in images. The YOLO model is seamlessly integrated into the OpenCV framework, enabling object detection. Post-processing techniques, such as non-maximum suppression, are used to modify detection results and improve accuracy. Visualizations, such as bounding boxes and labels, are used to help interpret the discovered items. The project finishes by investigating potential expansions and optimizations, such as custom dataset training and deployment on edge devices, opening up new paths for further investigation and development. This project provides developers with the tools and knowledge they need to build effective object detection systems for a wide range of applications, from surveillance and security to autonomous vehicles and augmented reality.


Vijay Verma

Binary Segmentation of PCB Components Using U-Net Model

When & Where:


Zoom Meeting

Degree Type:

MS Project Defense

Committee Members:

Sumaiya Shomaji, Chair
Tamzidul Hoque
Zijun Yao


Abstract

This project explores the adaptation of the U-Net convolutional neural network, renowned for its medical image segmentation prowess, to the analysis of Printed Circuit Boards (PCBs). By utilizing the Fine-Printed Circuit Board Image Collection (FPIC) dataset, we address key challenges in PCB inspection, such as the precise segmentation of complex components, handling class imbalances, and capturing minute details.

The U-Net model has been finely tuned with an encoding-decoding architecture, enhanced by convolutional layers, batch normalization, and dropout techniques to extract and reconstruct high-quality features from PCB images effectively. The Dice coefficient, used as the loss function, significantly improves boundary accuracy, and manages class diversity. Throughout extensive training and validation phases, the model has demonstrated superior performance metrics compared to traditional methods, making substantial advancements in automated PCB inspection.

During the rigorous training and validation stages, the U-Net model demonstrated excellent performance metrics, eclipsing traditional inspection methods. For capacitors, the model achieved a training accuracy of 95.03% and a validation accuracy of 95.92%. For resistors, training using transfer learning techniques resulted in even more remarkable performance, with training accuracy reaching 98% and validation accuracy hitting 98.23%. These metrics highlight the model's robustness and accuracy, marking a significant advancement in automated PCB inspection and suggesting the model's potential for wider industrial applications in multiclass component segmentation within complex PCB.


Ruturaj Vaidya

Exploring binary analysis techniques for security

When & Where:


Zoom

Degree Type:

PhD Dissertation Defense

Committee Members:

Prasad Kulkarni, Chair
Alex Bardas
Drew Davidson
Esam El-Araby
Michael Vitevitch

Abstract

In this dissertation our goal is to evaluate how the loss of information at binary-level affects the performance of existing compiler-level techniques in terms of both efficiency and effectiveness. Binary analysis is difficult, as most of semantic and syntactic information available at source-level gets lost during the compilation process. If the binary is stripped and/ or optimized, then it negatively affects the efficacy of binary analysis frameworks. Moreover, handwritten assembly, obfuscation, excessive indirect calls or jumps, etc. further degrade the accuracy of binary analysis. Challenges to precise binary analysis have implications on the effectiveness, accuracy, and performance, of security and program hardening techniques implemented at the binary level. While these challenges are well-known, their respective impacts on the effectiveness and performance of program hardening techniques are less well-studied.

In this dissertation, we employ classes of defense mechanisms to protect software from the most common software attacks, like buffer overflows and control flow attacks, to determine how this loss of program information at the binary-level affects the effectiveness and performance of defense mechanisms. Additionally, we aim to tackle an important problem of type recovery from binary executables that in turn help bolster the software protection mechanisms.


Jianpeng Li

BlackLitNetwork: Advancing Black Literature Discovery Through Modern Web Technologies

When & Where:


LEEP2, Room 1420

Degree Type:

MS Project Defense

Committee Members:

Drew Davidson, Chair
Sumaiya Shomaji
Han Wang


Abstract

Advancements in web technologies have significantly expanded access to diverse cultural narratives, yet black literature remains underrepresented in digital domains. The BlackLitNetwork addresses this oversight by harnessing Elasticsearch, MongoDB, React, Python, CSS, HTML, and Node.js, to enhance the discoverability and engagement with black novels. A major component of the platform is a novel generator built with Elasticsearch, which employs powerful full-text search capabilities, essential for users to navigate an extensive literary database effectively.

MongoDB supports the archives platform with a flexible data schema for managing varied literary content efficiently, while Python facilitates robust data cleaning and preprocessing to ensure data integrity and usability. The user interface, created using React, transforms Figma designs from our design team into a dynamic web presence, integrating HTML and CSS to ensure both aesthetic appeal and accessibility.

To further enhance security and manageability, we've implemented a Node.js backend. This layer acts as a middleware, managing and processing requests between our frontend and Elasticsearch. This not only secures our data interactions but also allows for request handling before querying Elasticsearch. This architecture ensures that BlackLitNetwork remains scalable and maintainable.

BlackLitNetwork also features specialized pages for podcasts, briefs, and interactive data visualizations, each designed to highlight historical, and contextual elements of black literature. These components aid in fostering a deeper understanding, establishing BlackLitNetwork as a tool for scholars. This project not only enriches the field of humanities but also promotes a broader understanding of the black literary heritage, making it a resource for researchers, educators, and readers keen on exploring the richness of black literature.


Aiden Liang

Enhancing Healthcare Resource Demand Forecasting Using Machine Learning: An Integrated Approach to Addressing Temporal Dynamics and External Influences

When & Where:


Nichols Hall, Room 246 (Executive Conference Room)

Degree Type:

MS Project Defense

Committee Members:

Prasad Kulkami, Chair
Fengjun Li
Zijun Yao


Abstract

This project aims to enhance predictive models for forecasting healthcare resource demand, particularly focusing on hospital bed occupancy and emergency room visits while considering external factors such as disease outbreaks and weather conditions. Utilizing a range of machine learning techniques, the research seeks to improve the accuracy and reliability of these forecasts, essential for optimizing healthcare resource management.

The project involves multiple phases, starting with the collection and preparation of historical data from public health databases and hospital records, enriched with external variables such as weather patterns and epidemiological data. Advanced feature engineering is key, transforming raw data into a machine learning-friendly format, including temporal and lag features to identify patterns and trends.

The study explores various machine learning methods, from traditional models like ARIMA to advanced techniques such as LSTM networks and GRU models, incorporating rigorous training and validation protocols to ensure robust performance. Model effectiveness is evaluated using metrics like MAE, RMSE, and MAPE, with a strong focus on model interpretability and explainability through techniques like SHAP and LIME. The project also addresses practical implementation challenges and ethical considerations, aiming to bridge academic research with practical healthcare applications. Findings are intended for dissemination through academic papers and conferences, ensuring that the models developed meet both the ethical standards and practical needs of the healthcare industry.


Thomas Atkins

Secure and Auditable Academic Collections via Hyperledger Fabric-Based Smart Contracts

When & Where:


Nichols Hall, Room 246

Degree Type:

MS Thesis Defense

Committee Members:

Drew Davidson, Chair
Fengjun Li
Bo Luo


Abstract

This paper introduces a novel approach to manage collections of artifacts through smart contract access control, rooted in on-chain role-based property-level access control. This smart contract facilitates the lifecycle of these artifacts including allowing for the creation, modification, removal, and historical auditing of the artifacts through both direct and suggested actions. This method introduces a collection object designed to store role privileges concerning state object properties. User roles are defined within an on-chain entity that maps users' signed identities to roles across different collections, enabling a single user to assume varying roles in distinct collections.

Unlike existing key-level endorsement mechanisms, this approach offers finer-grained privileges by defining them on a per-property basis, not at the key level. The outcome is a more flexible and fine-grained access control system seamlessly integrated into the smart contract itself, empowering administrators to manage access with precision and adaptability across diverse organizational contexts. This has the added benefit of allowing for the auditing of not only the history of the artifacts, but also for the permissions granted to the users. 


Theodore Harbison

Posting Passwords: How social media information can be leveraged in password guessing attacks

When & Where:


Zoom (ID: 7858648812, Pass: 348348

Degree Type:

MS Thesis Defense

Committee Members:

Hossein Saiedian, Chair
Fengjun Li
Heechul Yun


Abstract

The explosion of social media, while fostering connection, inadvertently exposes personal details that heighten password vulnerability. This thesis tackles this critical link, aiming to raise public awareness of the dangers of weak passwords and excessive online sharing. We introduce a novel password guessing algorithm, SocGuess, which capitalizes on the rich trove of information on social media profiles.

SocGuess leverages Named Entity Recognition (NER) to identify key data points within this information, such as dates, locations, and names. To further enhance its accuracy, SocGuess is trained on the rockyou dataset, a large collection of leaked passwords. By identifying different kinds of entities within these passwords, SocGuess can calculate the probability of these entities appearing in passwords.

Armed with this knowledge, SocGuess dynamically generates password guesses in order of probability by filling these entity placeholders with the corresponding data points harvested from the target’s social media profiles. This targeted approach shows SocGuess to crack 33% more passwords than existing algorithms during experimentation, demonstrably surpassing traditional methods.