137091 research outputs found
Sort by
Algebraic Methods in Spacecraft Navigation
No full textThe technological advancements of recent decades and the enduring drive to explore the universe are enabling new types of space missions, each presenting unique challenges. Deep space missions increasingly require autonomous navigation, for which Optical Navigation (OPNAV) offers a promising solution. As a result, developing and refining OPNAV algorithms is a vibrant area of research.
Algebraic geometry provides a convenient mathematical framework for OPNAV, as many objects and phenomena in spacecraft navigation can be described in polynomial form. Keplerian orbits, elliptical crater rims, ellipsoidal celestial bodies, atmospheric bands, planetary rings, are all curves and surfaces that can be modeled as degree-two polynomials. Additionally, phenomena and constraints such as the Doppler effect, stellar aberration, image distortion, and line-of-sight observations are naturally expressed in polynomial form.
This dissertation explores the application of techniques from algebraic and projective geometry to address navigation-related challenges. In the first part, analytical tools are leveraged to obtain the closed-form expression of a crater rim imaged with a pushbroom camera. Crater rims are often visible in the images captured by such sensors, and knowledge of their analytical shape enables interesting capabilities, such as crater reconstruction and spacecraft state estimation directly processing the image.
Subsequently, projective geometry is used to develop an analytical framework for parameterizing Keplerian orbits. This led to a novel velocity propagation technique, and to a solution to the problem of fitting a conic with constrained focus location to three points. Additionally, parameter homotopy continuation is used to initialize the state of an orbiting transmitter.
Finally, algebraic geometry is used to develop a technique for partially calibrating a camera from the image of a single celestial body. Then, the conic intersection problem is revisited and an algorithmic framework providing a simpler solution compared to current techniques is developed.Ph.D.Aerospace Engineerin
Harnessing Large Language Models Towards More Accessible Hardware Accelerator Design
No full textToday's computing landscape, increasingly dominated by computationally demanding algorithms such as deep learning, requires specialized hardware accelerators for efficiency. However, traditional hardware design processes rely heavily on expert-driven tasks like register-transfer-level (RTL) coding and optimization, resulting in lengthy development cycles and limited accessibility. This bottleneck widens the gap between rapid advancements in application domains and the comparatively slower evolution of supporting hardware. Meanwhile, Large Language Models (LLMs) are transforming technical domains like software engineering, by enabling the translation of natural language descriptions into functional code, raising the critical question: Can we similarly leverage LLMs to streamline hardware accelerator design into a flexible and broadly accessible process? Consequently, this thesis aims to reshape hardware design into a more automated and inclusive endeavor, allowing engineers, researchers, and students from diverse backgrounds to more easily and efficiently engage in hardware development tasks, including hardware coding, optimization, and debugging, through hardware-specialized AI assistants. Bridging LLM capabilities and hardware design practices could significantly democratize hardware development, shorten design iteration cycles, and enable hardware innovation to keep pace with rapidly evolving software and AI applications.
The aforementioned vision faces steep challenges. Despite vast training, current pretrained LLMs inherently lack expertise in specialized hardware semantics like Verilog, timing constraints, and synthesizability. Another critical hurdle is complexity management, as hardware design spans multiple stages—from high-level architecture to detailed implementation—necessitating frameworks enabling LLMs to approach problems step-by-step rather than attempting overwhelming tasks at once. Ensuring verification and correctness is also essential since hardware errors can lead to severe failures; thus, enabling LLM-generated designs to autonomously verify and self-correct their work presents significant difficulties. Additionally, the challenge of optimization and performance tuning means LLMs need quantitative feedback mechanisms to iteratively refine designs for timing, area, and power efficiency, integrating AI creativity with rigorous EDA analytics. Lastly, balancing generality versus specialization requires our methods to remain broadly applicable across various hardware accelerators yet adaptable enough to excel in specialized domains, such as AR/VR codec avatars, without sacrificing overall flexibility and usability.
To address these challenges, my thesis introduces a novel ecosystem leveraging LLMs to democratize and automate hardware accelerator design. The core idea is enabling LLMs to collaboratively tackle complex hardware design tasks through systematic task planning and decomposition, generation, verification, and iterative refinement processes. Specifically, the proposed ecosystem comprises three hierarchical workflow levels, with concrete demonstrations of system integration and optimization:
1. Design Planning and Generation: We developed GPT4AIGChip, a structured framework that equips LLMs to effectively plan for hardware designs tasks. This is achieved by breaking down complex tasks into well-defined stages, including architectural planning, module decomposition, and code generation. This modular approach ensures that resulting hardware designs remain manageable, contextually coherent, and practically viable for real-world applications. This level shows that LLMs, with structured guidance and training, can reliably produce complex hardware designs, marking a significant paradigm shift in leveraging more flexible format, such as natural language, for precise engineering tasks. 2. Hardware Knowledge Enhancement: Recognizing that existing LLMs often lack specific hardware design knowledge, we introduced MG-Verilog, an automated framework designed to generate multi-grained synthetic training data tailored for hardware design. MG-Verilog produces high-quality hardware design datasets through a novel Pyramid of Thoughts (PoT) data structure, enhanced by curated Retrieval-Augmented Generation (RAG) techniques, and automated validation. This structured dataset significantly enhances LLMs' capabilities, effectively addressing critical knowledge gaps across various hardware design scenarios, and substantially improves their performance in practical hardware design contexts through methods like fine-tuning and in-context learning. 3. Verification and Iterative Improvement: To ensure functional correctness and facilitate iterative design enhancements, we developed SLAVA, a scalable LLM-driven framework that introduces an assertion-guided automated self-refinement loop integrating RTL generation, verification, and iterative repair. Unlike earlier frameworks that focused on isolated assertion generation, SLAVA autonomously generates modular RTL, derives verification plans and SystemVerilog Assertions (SVAs), and leverages verification feedback to iteratively refine both RTL and SVAs. SLAVA intelligently analyzes verification outcomes to determine whether issues stem from genuine design errors or verification artifacts, employing advanced techniques such as systematic SVA source attribution, graph-based test-failure localization, and context-aware repair strategies. This comprehensive approach significantly enhances the robustness and accuracy of automated design refinement, effectively bridging the gap between design and verification stages. Beyond these three hierarchical workflow levels, this thesis also presents a comprehensive demonstration of practical system-level integration and optimization: To provide practical integration and optimizations of the LLM-driven for hardware design ecosystem, we introduced AutoAI2C and Re-CATA frameworks. AutoAI2C merges traditional hardware automation techniques and tool-chains with AI-driven approaches to optimize accelerator designs, focusing on real-world efficiency. Concurrently, Re-CATA showcases these integrated methodologies within a customizable, domain-specific accelerator tailored to AR/VR applications, illustrating the flexibility and applicability of the ecosystem in realistic scenarios. This practical demonstration provides a glimpse into a future where engineers collaborate seamlessly with AI to streamline and innovate hardware design.
A central theme across these levels is integration, where each level collaboratively complementing the others, enhancing LLM knowledge, generating hardware designs, verifying and refining these designs, optimizing them through conventional tools, and ultimately tailoring outputs for practical applications. In the long term, these techniques aim to transform hardware development, making it as rapid, iterative, and accessible as software development. This transformation could result in a highly adaptive hardware design process, enabling a broader range of developers to efficiently create specialized hardware tailored specifically to their application needs. Although ambitious, the work presented in this thesis represents a significant step toward achieving this vision of automated and accelerated hardware design. Last but least, rather than replacing human engineers, this approach seeks to elevate their roles to supervisory and creative positions, delegating routine tasks to AI. If successfully implemented, this synergy could drive greater innovation, shifting the primary challenge from technical execution to idea generation, allowing engineers to articulate objectives while AI autonomously produces optimized hardware solutions.Ph.D.Computer Scienc
Energy-Efficient Hardware Acceleration of Transformer-Based Models
No full textThis research presents a software-hardware co-optimization framework for energy-efficient deployment of transformer models on FPGAs. It introduces novel quantization techniques tailored for both BERT and generative LLMs. For weight quantization, the proposed Hessian-based parameter-wise method assigns optimal bit precision based on sensitivity analysis, while the proposed row-wise quantization enhances hardware efficiency by converting mixed-precision matrices into two uniform-precision blocks. For attention activations, the proposed Weight-Hessian-aware KV cache quantization applies intra-layer mixed-precision using precomputed sensitivities, eliminating runtime overhead. To further improve hardware efficiency, the proposed Query-Key coupled scheme aligns bit precision within each outer product pair, reducing implementation complexity. A concurrent quantization approach jointly optimizes row-wise weight and Query-Key activation precision using multi-precision formats, improving both compression and energy efficiency. These techniques are implemented on a novel multi-precision FPGA accelerator for BERT and GPT-2, capable of handling both power-of-two and non-power-of-two bit-widths. With optimized dataflow, the design minimizes off-chip memory access and significantly outperforms prior solutions in both energy efficiency and inference performance.Ph.D.Electrical and Computer Engineerin
Approximation and Optimization Approaches for Graph Cuts, Experimental Design, and Routing Problems
No full textThis dissertation explores three topics: maximum graph cuts, the design of experiments, and the traveling salesman problem. First, we study MAX-3-SECTION, where the goal is to partition a weighted graph into three equally sized parts to maximize the weight of crossing edges. We present a new approximation algorithm that improves upon the best-known factor by rounding a semidefinite programming relaxation. The second topic focuses on design of experiments, where the objective is to select a limited subset of experiments from a universe to maximize the information gained about an underlying system. This is typically achieved by optimizing criteria such as D-optimality or A-optimality. A major challenge arises when the universe of experiments is too large to be explicitly enumerated. We address this challenge in two different settings. In the first work, we consider the case where the experiment space consists of the boolean hypercube, with D-optimality as the objective. We show that the well-known Federov exchange algorithm reduces to solving a quadratic optimization problem that generalizes MAX-CUT. We include a theoretical analysis that applies approximation algorithms for MAX-CUT similar to those used in the first part of the thesis. In the second
work, we consider the case where the experiment space is the surface of the unit sphere, with A-optimality as the objective. We show that the Federov exchange algorithm reduces to solving a Rayleigh quotient problem. Both works include an experimental study demonstrating the scalability of our approach. Finally, the third part is about a variant of the traveling salesman problem (TSP) called multi-visit TSP. In multi-visit TSP, each city has a visit request r(v) and the goal is for a salesman to go on a tour so that each city is visited r(v) times. In this work, we show that a LP relaxation based algorithm for TSP can be converted to a LP relaxation based algorithm for multi-visit TSP with the same approximation factor. We apply our reduction to several variants of multi-visit TSP to either improve the current best approximation factors.Ph.D.Algorithms, Combinatorics, and Optimizatio
Heparin-Based Hydrogel/3D Printed Scaffold Composites for Cartilage Regeneration
No full textCartilage is a specialized connective tissue known for maintaining the structural integrity and functionality of various body parts, including joints, the nose, and the ears. Unlike most other tissues in the body, cartilage exhibits limited self-repair capabilities due to its avascular nature and low cell turnover rate. Traditional approaches, including autologous cartilage grafts and prosthetic devices, are associated with inherent limitations such as donor site morbidity, donor tissue availability, and suboptimal aesthetic outcomes. Moreover, the complex three-dimensional (3D) architecture of the ear, comprising intricate contours and curves, further complicates the restoration process. As such, there is a compelling clinical need for innovative auricular reconstruction strategies capable of addressing these challenges while achieving durable and aesthetically pleasing outcomes. This thesis work aimed to streamline cartilage tissue engineering by developing a novel approach for in situ assembly of bioactive composite structures in the operating room (OR), circumventing regulatory challenges associated with extensive in vitro chondrocyte culturing. Composite scaffolds capable of both biological stimulation and structural support essential for cartilage regeneration were developed in this thesis. In Aim 1, novel 3D printed-hydrogel composite scaffolds were developed and characterized. In Aim 2, the impact of cartilage digestion level on extracellular matrix production was assessed by encapsulating isolated chondrocytes, partially digested cartilage, and minced cartilage pieces in PEGDA-DTT hydrogels. Finally, in Aim 3, chondrocyte maintenance and cartilaginous matrix production were explored by encapsulating chondrocytes within various levels of heparin-sulfated hydrogels.Ph.D.Biomedical Engineerin
Modeling Pseudomonas aeruginosa Aggregates: A Multi-Scale Simulation Approach to Understanding Growth Dynamics
No full textPseudomonas aeruginosa is a notorious opportunistic pathogen that forms aggregates, which play a crucial role in its pathogenesis and antibiotic resistance. Despite their significance, the growth dynamics of these aggregates remain unclear, hindering the development of effective therapeutic strategies. Most existing simulations remain focused on biofilm growth and maintenance, ignoring the dynamics of suspended aggregates. This dissertation presents a multi-scale agent-based simulation approach to model the growth dynamics of P. aeruginosa aggregates. We develop a comprehensive framework that bridges the gap between cell-level interactions and aggregate behavior, balancing biological accuracy with computational efficiency. The simulation aims to capture the interactions between bacterial motility, adhesion, nutrient availability and spatial limitations in shaping aggregate growth. This work has significant implications for the design of novel antibacterial therapies and the development of predictive models for P. aeruginosa infections.UndergraduateComputer Scienc
A Lived Experience: A Journey Through Alcohol Use Disorder
No full textThis short documentary was created as a course requirement in HTS 3086 – Sociology of Medicine and Health under the supervision of Dr. Jennifer Singh.Runtime: 05:58 minutesThis documentary explores the illness experience of recovering from Alcohol Use Disorder through Alcoholics Anonymous
Analytics of Joint Logistics for Collecting Used Battery
No full textOur group planed used battery recycling. We succeeded in our recycling a dry cell, and making ferrite powder. We also suppose the collecting method for used batteries. We simulated TSP method of Genetic Algorithm. But the calculation has the limit to solve. We suggest the new method to solve the theme. The new method has the peculiar point like not only shorter route but also maximum collection. We discus those points below
Developing meta'omic methods for forensic analysis of wastewater infrastructure performance
No full textWastewater infrastructure represents the components of the bult environment which capture, convey, and treat wastewater. The type of wastewater infrastructure employed can vary globally, based on social and economic conditions. Yet, all wastewater infrastructure is replete with a diverse array of microbial life. The communities formed by these microbes play key roles in biological wastewater treatment. However, microorganisms can present substantial hazards when released into the environment since they often harbor human pathogens, antimicrobial resistance genes, and other pollutants. Yet, global estimates for the percentage of wastewater escaping proper treatment exceed 80%. Knowledge of when and why failure occurs among wastewater infrastructure systems is essential for improving such statistics. Expanding forensic tools available to environmental engineers for interrogating the performance of wastewater infrastructure, from conveyance in collection systems to biological treatment at WWTPs, will be essential to meet sanitation and sustainable development goals now and in the future. The application of shotgun-based nucleic acid sequencing, i.e., meta’omic methodologies, to examine the root cause of failure in these systems remains rare, in part because of several limitations ranging from uncertain analytical limits, lack of population proportioning benchmarking, and uncharted microbial genetic and functional diversity. Yet these data streams promise to assist environmental engineering forensics and this dissertation explores the development of meta’omic approaches for forensic analysis of wastewater infrastructure.
This dissertation begins with the development of a theoretical basis in support of using metagenomic sequencing to track microbial populations in food and environmental samples (Chapter 2). This framework proposed multiple avenues for estimating a key parameter for microbial tracking: the detection limit. The framework was based on a target feature’s sequence characteristics and methods for calculating abundance a priori and a posteriori based on metagenome read mapping. Utilizing this theoretical framework, a metagenomic theoretical limit of detection (thLOD) was tested using simulated sewage spills in laboratory mesocosms combining sequencing, culture-based, and qPCR-based data (Chapter 3). Sewage-associated populations were tractable and quantifiable across 4-7 days of monitoring and in silico experiments validated the proposed thLOD. Chapter 3 also explored the use of shotgun metagenomic data to perform source apportionment – this is, to describe the fraction of fecal contamination belonging to a particular source. A major limitation uncovered by these efforts was the need to understand which microbial populations were specific or cross-reactive.
Accordingly, a whole genome database was curated for fecally shed prokaryotes across multiple common contaminating sources to overcome this issue (Chapter 4). Comparative genomics revealed that the species level performed well as a cutoff for grouping populations into source-specific categories though substantial sequencing efforts are still needed for many hosts, especially ruminants. To combine these source curated databases with a population proportioning pipeline, a new bioinformatic tool was developed as part of this thesis called “SourceApp” (Chapter 5). The tool was then tested in a double-blind study using mesocosms of known composition to benchmark its source attribution, apportionment, and cell fractioning capabilities. Extensive parameter tuning was investigated to identify parameters of best choice for these efforts. The findings of this portion of the thesis position metagenomic based fecal source tracking (FST) as a powerful technique for source apportionment – which remains difficult with existing techniques. Additionally, the development of SourceApp should accelerate application of metagenomic FST more broadly and aid with reproducibility across future work.
Lastly, this thesis applied the methodologies developed for FST to tracking functional microbial populations in a continuous mixed culture environmental bioprocess for enhanced biological phosphorus removal (EBPR) (Chapter 6). Key functional populations were identified based on taxonomic classification and gene annotations. Changes in the relative abundances of these populations corresponded with periods of decreased EBPR performance. Statistical modeling of process performance was then used to implicate the importance of flanking populations which were not identified as putative functional populations but may play important roles in process performance. Follow up efforts combining metagenomic and metatranscriptomic sequencing identified specific genes carried by functional and flanking populations which were up and down regulated during process upsets. These genomes and the genes identified through metatranscriptomic inquiry can serve as biomarkers for ongoing efforts to probe the performance of full-scale EBPR with more rapid, real-time quantitative methods (e.g., RT-ddPCR).Ph.D.Environmental Engineerin