Theses and Dissertations

Item

Scalable, Efficient, and Fair Algorithms for Structured Convex Optimization Problems

(Georgia Institute of Technology, 2023-08-24) Ghadiri, Mehrdad ; Vempala, Santosh S. ; Peng, Richard ; Singh, Mohit ; Brand, Jan van den ; Gupta, Swati ; Computer Science

The growth of machine learning and data science has necessitated the development of provably fast and scalable algorithms that incorporate ethical requirements. In this thesis, we present algorithms for fundamental optimization algorithms with theoretical guarantees on approximation quality and running time. We analyze the bit complexity and stability of efficient algorithms for problems including linear regression, $p$-norm regression, and linear programming by showing that a common subroutine, inverse maintenance, is backward stable and that iterative approaches for solving constrained weighted regression problems can be carried out with bounded-error pre-conditioners. We also present conjectures regarding the running time of computing symmetric factorizations for Hankel matrices that imply faster-than-matrix-multiplication time algorithms for solving sparse poly-conditioned linear programs. We present the first subquadratic algorithm for solving the Kronecker regression problem, which improves the running time of all steps of the alternating least squares algorithm for the Tucker decomposition of tensors. In addition, we introduce the Tucker packing problem for computing an approximately optimal core shape for the Tucker decomposition problem. We prove this problem is NP-hard and provide polynomial-time approximation schemes for it. Finally, we show that the popular $k$-means clustering algorithm (Lloyd's heuristic) can result in outcomes that are unfavorable to subgroups of data. We introduce the socially fair $k$-means problem for which we provide a very efficient and practical heuristic. For the more general problem of $(\ell_p,k)$-clustering problem, we provide bicriteria constant-factor approximation algorithms. Many of our algorithms improve the state-of-the-art in practice.

Item

Discovery and Development Strategies of Combination Nanomedicines for Childhood Blood Cancers

(Georgia Institute of Technology, 2023-04-17) Kelvin, James Michael ; Dreaden, Erik C. ; DeRyckere, Deborah ; Kemp, Melissa L. ; Du, Yuhong ; Xia, Younan ; Biomedical Engineering (Joint GT/Emory Department)

Blood cancers are the most frequently diagnosed and the second deadliest of malignancies in children. Despite advances in multiagent chemotherapy that have contributed to improved survival rates, nearly half of patients who survive will suffer from treatment associated long-term toxicities, and a considerable number of patients eventually relapse with poor survival prognoses thereafter. Thus, there is an urgent and unmet clinical need to develop novel therapies that improve treatment outcomes for pediatric patients with leukemia. This Dissertation describes the discovery and development of combination nanomedicines to address this need. We integrate three distinct treatment strategies to maximize the therapeutic potential of novel drug combinations: (i) use tyrosine kinase inhibition to exploit ectopic molecular vulnerabilities; (ii) identify synergistic drug ratios that amplify tumor cell killing; and (iii) formulate therapeutic combinations in liposomal nanocarriers for intracellular delivery of constitutively synergistic drug ratios. Governing our approach is the hypothesis that the conditional delivery of synergistic drug ratios identified in vitro will result in reduced disease burden and prolonged survival in mouse models of leukemia when compared to additive or antagonistic nanoformulations. Our efforts to discover and develop synergistic nanomedicines are mirrored between studies of pediatric acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML). We begin by describing results from a novel, combinatorial high-throughput drug screen in which we identified ratio-dependent synergy between a dual MERTK/FLT3 inhibitor, MRX-2843, and cytotoxic chemotherapy (vincristine) in T-cell ALL (T-ALL) or BCL-2 inhibition (venetoclax) in AML. We then used computational models to select optimal pairwise drug combinations that exhibited robust inhibition of cell expansion and conserved ratiometric synergy in T-ALL and AML lineages. We characterized pairwise drug synergy by building predictive classifiers of drug responses between MRX-2843 and venetoclax in AML cell lines, and used RNA sequencing to explore functional ontologies that undergird the mechanism of drug synergy between MRX-2843 and vincristine in T-ALL. Next, we developed a clinical-scale manufacturing method that (co-)encapsulated defined drug ratios in liposomal nanoparticles. Nanoformulations delivered intracellular drug ratios at defined stoichiometries and demonstrated synergistic activity in primary patient samples—consistent with high-throughput screens—such that nanoparticles magnified dose-dependent synergy relative to matched free drug ratios. Finally, we directly compared synergistic, additive, and antagonistic nanomedicines in a mouse model of early thymic precursor ALL (ETP-ALL). We found that increasing the dose of MRX-2843 sensitized ETP-ALL cells to vincristine chemotherapy in vivo, and that synergistic and additive nanoformulations reduced disease burden and extended survival relative to liposomal controls. Counter to the hypothesis, the additive nanoformulation most effectively controlled disease and extended survival, a finding that contextualizes the prioritization of ratiometric synergy and therapeutic efficacy in nanomedicine design. In sum, we present a systematic approach to combination drug discovery and development for novel nanomedicines in pediatric leukemias. Our findings underscore the clinical relevance of MRX-2843 in combination with venetoclax in pediatric AML and support the translation of co-formulated MRX-2843 and vincristine nanomedicines for the treatment of pediatric patients with T-ALL. Our generalizable approach may be applied to different drug combinations that treat hematologic neoplasms or other cancers.

Item

Development of Additively Manufactured Molybdenum and Improvement via Lanthanum Oxide Addition

(Georgia Institute of Technology, 2023-08-30) Hutchinson, Andrew ; Stebner, Aaron ; Kurfess, Thomas ; Neu, Richard ; Curran, David ; Mechanical Engineering

Additive manufacturing (AM) technologies have provided an avenue for processing traditionally difficult to manufacture metals. Manufacturability of one such material, molybdenum, has remained on the forefront of challenges hindering its widespread application. In fact, the current manufacturing methods employed by Framatome to create sintering boats for fissionable fuel out of a molybdenum-lanthanum oxide alloy often result in residual stress buildup, defects, and subsequent part failure. Thus, the use of AM methods could provide a promising path to remediate the expensive operations and tooling needed to manufacture molybdenum parts. To date, work done on the additive manufacturing of molybdenum has focused on small scale parts with processes such as powder bed fusion. Novel investigation of both pure molybdenum and molybdenum alloyed with nanoparticle lanthanum oxide manufactured by a directed energy deposition – laser beam – powder blown (DED-LB-PB) AM process is presented in this work. Importantly, the discovery of process parameter sets corresponding to high densities of molybdenum is accomplished via response surface methodology experimental design. Maximum densities achieved are 96.99% and 99.87% in the pure molybdenum and alloyed molybdenum systems, respectively, thus demonstrating the capability of the DEDLB-PB method for manufacturing commercial parts with future work. Furthermore, microstructural characterization of the AM specimens produced has demonstrated the effectiveness of the nanoparticle lanthanum oxide addition to reduce grain size, reducing millimeter-tall grains to hundreds of microns. Accordingly, grain boundary cracking is reduced significantly, allowing for the creation of larger mechanical samples. The compression testing of these alloyed samples yielded an average strength of 217 MPa, further indicating the possibility of commercial part manufacturability. Chemical analysis data alluded to the loss of lanthana during the DED-LB-PB process. Dimensional stability and accuracy of parts made with the AM method showed relationships to varied parameters of laser power, scan speed, and mass flow, as well as to the addition of lanthana.

Item

Methodologies for Modeling and Optimization of 2.5-D and 3-D Integration Architectures for Compute-In-Memory Applications

(Georgia Institute of Technology, 2023-08-28) Kaul, Ankit ; Bakir, Muhannad S. ; Raychowdhury, Arijit ; Naeemi, Azad ; Datta, Suman ; Smet, Vanessa ; Dhavaleswarapu, Hemanth ; Electrical and Computer Engineering

The objective of this research is to investigate power delivery network (PDN) and thermal management constraints in emerging 3-D heterogeneous integration (HI) architectures for compute-in-memory (CIM) applications. First, design trade-offs in the PDN of bridge-chip based 2.5-D heterogeneous platforms are investigated. It is demonstrated that including a PDN in the bridge-chip can provide significant reduction in DC-IR drop, Ldi/dt noise, and high-frequency ripple compared to the baseline. Second, a comprehensive design-space exploration of PDN design for 3-D-HI CIM hardware is presented. A methodology is proposed to evaluate and quantify trade-offs between power delivery design parameters and CIM performance metrics. Subsequently, a device-integration methodology is proposed to quantify the thermal-driven impact of integration architectures on resistive random-access memory (RRAM) reliability for CIM applications. Two 3-D-HI accelerator designs are benchmarked against monolithic 2-D and balanced integration design parameters are reported. Finally, a back-end-of-line (BEOL)-embedded chiplet integration architecture (polylithic 3-D) is proposed. Polylithic 3-D integration represents a densely integrated system divided into multiple device tiers where custom chiplets can be embedded into the back end of a primary tier with extremely efficient signaling and large bandwidth density. Design optimization strategies for PDN and thermal management in polylithic 3-D integration are presented and benchmarked against conventional 3-D integration. The potential impact of this research and potential future directions are summarized.

Item

Rhythm Recreation Study To Inform Intelligent Pedagogy Systems

(Georgia Institute of Technology, 2023-08-28) Alben, Noel ; Condit-Schultz, Nat ; Freeman, Jason ; Brown, Thackery ; Music

Web-based intelligent pedagogy systems have great potential to provide interactive music lessons to those unable to access conventional, face-to-face music instruction from human experts. A key component of any effective pedagogy system is the expert domain knowledge used to generate, present, and evaluate the teachable content that makes up the ''syllabus'' of the system (Brusilovskiy, 1994). In this work, we investigate the application of computational musicology algorithms to devise the ''syllabus'' of intelligent rhythm pedagogy software. Many computational metrics that quantify and characterize rhythmic patterns have been proposed (Toussaint). We employ Cao et al.'s (2012) family theory of rhythms as a metric of rhythmic similarity and an entropy-based coded-element metric of rhythmic complexity (Thul, 2008). Both metrics have been shown to correlate with human judgments of rhythmic similarity and complexity. A rhythmic syllabus that uses these metrics to determine the order in which rhythmic patterns are learned will be easier for musicians to progress through. We test this hypothesis in a rhythm reproduction study hosted on a custom-designed web-based experimental interface. Our experiment consists of six individual blocks: In each block, a participant listens to five unique rhythmic patterns, which they must then reproduce by clapping into their computer's microphone. Each rhythmic pattern is two measures long on an eighth-note grid, presented at 105 BPM, and looped four times. The order and content of rhythmic patterns within each block are determined using our chosen complexity and similarity metrics. A participant completes a block when they reproduce all the rhythmic patterns of the block within the performance constraints defined by automatic performance assessment built into the experimental interface. Each of our six blocks represents key interactions: the order of the stimuli determined by our prescribed metrics, melodic information added to the rhythmic stimuli, and the presence of a visual representation of the rhythmic pattern. We also have control blocks where the patterns of each block are selected randomly without any theoretically informed metrics. Dependent variables to measure the effectiveness of the syllabus are the number of trials taken to reproduce a given rhythmic stimuli accurately. Participant reproductions are stored to afford future analyses, and the designed interface helps efficiently automate the data collection, making it more accessible for future rhythm reproduction studies. We conducted the rhythm recreation study with 28 participants across the United States, who accessed the experiment through a web-based portal. The data gathered from our experiment implies that computational music theory algorithms can contribute to creating syllabi that align with human perception. However, these results deviate from my initial predictions. Furthermore, It appears that while incorporating visual stimuli aided in learning rhythmic patterns, the introduction of pitched onsets negatively affected reproduction performance.

Item

Device-level thermal management and reliability of gallium nitride and aluminum gallium nitride high electron mobility transistors

(Georgia Institute of Technology, 2023-08-24) Hines, Nicholas J. ; Graham, Samuel ; Kumar, Satish ; Smet, Vanessa ; Choi, Sukwon ; Klein, Brianna A. ; Mechanical Engineering

The fields of power and radio frequency (RF) electronics have experienced tremendous growth over recent years as gallium nitride (GaN) device technology is maturing. GaN high electron mobility transistors (HEMTs) are particularly well-suited for high-power and high- frequency applications due to their excellent sheet charge density and channel mobility, and the large bandgap energy of GaN. However, GaN HEMTs suffer from acute self-heating that limits their performance in high-power and high-frequency applications. The most recent advancements in GaN HEMT device-level thermal management consist of integrating high-thermal conductivity CVD diamond substrates to GaN HEMT device layers (GaN-on-diamond technology). While the thermal merits for CVD diamond substrate integration are clear, the structural integrity and reliability of GaN-on-diamond HEMTs requires further investigation. To study the structural impact that CVD diamond integration has on GaN HEMTs, GaN-on-diamond materials fabricated by various techniques have been examined via optical stress metrology techniques. Ultra-wide bandgap (UWBG) aluminum gallium nitride (AlGaN) HEMTs have the potential to exceed the performance limitations of GaN HEMTs for the next generation of power and RF electronic device technologies. The acute self-heating challenges for high-power GaN HEMTs are exacerbated for AlGaN HEMTs because the thermal conductivity of AlGaN is an order of magnitude lower than that of GaN. The low thermal conductivity of AlGaN increases the device thermal resistance of AlGaN HEMTs and changes the transient thermal dynamics of AlGaN HEMTs under pulsed-mode operation. Therefore, AlGaN HEMT devices require novel device- level thermal management solutions to realize their theoretical performance potential. To address the thermal management challenges, novel device-level thermal management approaches have been identified via thermal finite element analysis (FEA) and in situ junction temperature experiments.

Item

Experimental study on the nonlinear mixing of ultrasonic waves in concrete using array technique

(Georgia Institute of Technology, 2023-08-21) Weiss, Fiona Jacqueline ; Jacobs, Laurence J. ; Kim, Jin-Yeon ; Kurtis, Kimberly E. ; Civil and Environmental Engineering

This research develops a procedure that combines array technology with non-collinear ultrasonic wave mixing to detect and scan internal microscale damage in a concrete prism specimen. By mixing two wave fronts of incident shear waves generated by two ultrasonic transducer arrays, one can exploit the underlying mechanics of nonlinear wave mixing to create a longitudinal mixed wave and measure the magnitude of this nonlinear wave at a frequency that is the sum of the fundamental frequencies. The frequency of the incident waves is chosen such that it is low enough to propagate without being scattered by the in- herently inhomogeneous concrete microstructure, while the resulting nonlinear phenomena are still sensitive to damage much smaller than the wavelength of the incident waves. The arrays enable beam steering, making it possible to scan for damage along an arc. Overall, scanning and imaging at different locations in a large volume throughout the specimen’s thickness is accomplished by manually adjusting the placement of the two arrays to move the mixing zone any desired, internal depth, while beam steering is used to scan at different locations of the same depth close to each other. The effectiveness of the proposed technique is demonstrated by characterizing different types of damage embedded at known locations in a concrete prism specimen. The results of this thesis are in accordance with previous research and show that beam steering along an arc to scan for damage in the concrete specimen is in fact possible.

Item

Understanding Failure Modes and High Load Operation in Compression-Ignition Engines With Ducted Fuel Injection

(Georgia Institute of Technology, 2023-08-14) Yraguen, Boni F. ; Steinberg, Adam M. ; Lieuwen, Timothy ; Ranjan, Devesh ; Mueller, Charles ; Sun, Wenting ; Mechanical Engineering

This thesis explores the potential of Ducted Fuel Injection (DFI) as a combustion strategy for reducing emissions from compression ignition (CI) engines. The research investigates the performance, failure modes, and optimization possibilities of DFI through experimental studies conducted in a pressure vessel and an optically accessible engine. First, a pressure vessel experiment is conducted to examine the impact of duct geometries on DFI performance. The analysis of duct diameter and length provides insights into their influence on combustion behavior. These findings are valuable for optimizing the design and configuration of the ducts. Additionally, a potential failure mode related to soot flare-ups is identified. These flare-ups manifest as sudden and inconsistent increases in soot luminosity. While combustion vessel experiments allow for flexibility in testing various duct configurations, they cannot fully capture the complexities of engine operation. To address this limitation, a parametric evaluation is carried out in an optical engine using two different duct configurations. The results of this study provide data on emissions, efficiency, and combustion behavior trends for different duct setups. The findings highlight the trade-offs between reducing emissions and improving fuel-conversion efficiency, emphasizing the need for careful selection of duct configuration. Notably, the presence of soot flare-ups is also observed in the engine. The further characterization of late-cycle flare-up behavior in the optical engine identifies distinct combustion modes associated with flare-ups and determines the factors influencing their occurrence. These insights help unravel the mechanisms behind this potential failure mode and guide strategies to mitigate its impact. This investigation advances the understanding of DFI for CI engines. It provides new data on DFI performance, identifies geometric parameter trends, characterizes failure modes, and offers insights for optimizing DFI implementation. This thesis lays the groundwork for further advancements in DFI technology and its practical application in achieving cleaner and more efficient combustion engines.

Item

Creating mobile manipulation capabilities: on dynamic modeling and control of Euler-Lagrange passive nonholonomic systems

(Georgia Institute of Technology, 2023-08-08) Aguilera Marinovic, Sergio Francisco ; Hutchinson, Seth ; Zhao, Ye ; Rogers, Jonathan ; Kemp, Charlie ; Dellaert, Frank ; Electrical and Computer Engineering

Mobile Manipulators (MM) have been introduced into various fields, from agriculture and warehouse automation to hospitals and personal assistants. Commonly used for fetching tasks, the range of capabilities that MM currently have is limited to traversing the environment and interacting with light objects. The goal of this thesis is to expand their capabilities by enabling MM to interact with and control heavy objects around the environment. Considering that large and heavy objects have nontrivial dynamics and momentum, this requires understanding how they behave and react to the interactions with the MM. From pushing wagons and trolleys in a warehouse, wheeled beds and wheelchairs in a hospital, and shopping carts, luggage, and strollers as personal assistant robots, there is a wide range of tasks where MMs push objects with changing inertial parameters. This thesis proposes a framework for interacting with and controlling wheeled objects with a wide range of inertial parameters. We study both the MM and object as two individual systems. For the object, we present a general dynamic model for various cart-like systems with unknown parameters, when the actuation is an external force applied by the MM. First, we present an adaptive control approach to tackle the system identification of the object and control for trajectory tracking. Then, we propose a position/force control scheme for the MM's manipulator to move the base along with the object while applying the required force at a given contact point on the object. Experimental results using a MM and a shopping cart are used to demonstrate the accuracy of the framework.

Item

Assessment and Analysis of Turbulent Flame Speed Measurements of Hydrogen-Containing Fuels

(Georgia Institute of Technology, 2023-08-14) Johnson, Henderson ; Lieuwen, Timothy C. ; Seitzman, Jerry ; Steinberg, Adam ; Sun, Wenting ; Blunck, David ; Aerospace Engineering

Global efforts to reduce greenhouse gas emissions and achieve a carbon-neutral economy have spurred the exploration of integrating hydrogen into various aspects of the global energy infrastructure. This can involve incorporating hydrogen into existing power generation applications or utilizing fuels with significant hydrogen content, such as syngas. However, the introduction of hydrogen poses significant challenges due to its potential to greatly impact the combustion process, with many aspects of its behavior not yet fully understood under practical gas turbine operating conditions. This thesis aims to investigate the influence of thermodynamic, fluid mechanic, and fuel factors on the turbulent global consumption speed, ST,GC, across different fuel types containing up to 90% hydrogen. This parameter represents the average rate of conversion of reactants to products relative to a specific iso-surface. The presented database encompasses three distinct fuel types: H2/CO, H2/CO/CH4/N2, and H2/CH4¬, which represent fuels that are either commonly encountered in practical applications or are of interest for future applications. The latter two fuels are new to the overall Georgia Tech database of turbulent flame speed measurements which increase the amount of high pressure data (up to 20 atm), and add data at preheat temperatures up to 500 K. The addition of this data is of great importance as it allows for further exploration of thermodynamic and fuel effects on ST,GC¬. The analysis of this database reveals several key findings. Firstly, regardless of whether the unstretched laminar flame speed, SL,0, is held constant, higher pressures lead to an increase in ST,GC across all fuel types. The preheat temperature is also shown to increase ST,GC, but when normalized by the laminar flame speed, it demonstrates a decrease. Moreover, the effects of hydrogen addition in H2/CO and H2/CO/CH4/N2 fuel blends are more pronounced compared to those in H2/CH4 fuels. Building upon prior studies that link these observations to mixture stretch sensitivity, the database is analyzed within the framework of a quasi-steady leading points concept model. In this framework, the maximum stretched laminar flame speed, SL,max, serves as the normalizing parameter. This approach proves effective for the H2/CO fuels discussed in this work, as it captures fuel effects at a fixed pressure and preheat temperature. However, a notable limitation arises in its inability to account for systematic differences in pressure and preheat temperature, indicating the need for a second correlating parameter. To identify this second parameter, a systematic investigation of three additional dimensionless numbers, namely the turbulent Reynolds number, Ret, time scale ratio, and acceleration ratio, is presented. Each of these numbers represents a different physical phenomenon that could potentially account for the observed variation in the data reported. The addition of Ret was considered in prior work; however, we identify that is insufficient as an appropriate scaling number due to its inconsistent correlation with preheat temperature. The acceleration ratio was introduced as a novel means of attempting to capture the ability of a flame to accelerate relative to the flow field. Similar to the Reynolds number, this approach showed limited ability to capture both pressure and preheat temperature effects; nevertheless, it does offer a new way to think about turbulence-flame interactions. Ultimately, the time scale ratio emerges as the optimal second correlating parameter due to its lesser degree of scatter compared to the acceleration ratio. This finding is significant, as it aligns with prior analyses that incorporated the time scale ratio to quantify non-quasi-steady chemistry effects at the leading point and demonstrates its promise as an appropriate scaling approach across a wide variety of conditions.

Theses and Dissertations

Permanent URI for this collection

Search within this collection.

Browse

Georgia Tech Library

Browse

Recent Submissions