137091 research outputs found
Sort by
Enhancing SSD Performance for GPUDirect Storage Systems Through Dynamic Address Allocation and Fine-Grained Address Mapping
No full textThe recent boom in machine learning (ML) and high performance computing (HPC) has necessitated efficient data processing to keep up with the increased rate of innovation. New machine learning and scientific computing models require larger datasets year-over-year, and this motivates the optimization of the data processing path related to a system’s storage. Furthermore, research has been increasingly using GPUDirectStorage systems to eliminate the CPU as an unnecessary middleman in the data processing involved in several applications. In this research, we examine several optimizations related to improving SSD performance within these GPUDirectStorage systems including fine-grained address mapping and token-based garbage collection. From our results, we found that these changes, alongside other optimizations, resulted in higher Input/Output Operations per Second (IOPS) and minimized tail latencies, overall increasing the efficiency of this type of architecture. Fine-grained address mapping enables write operations being done at a smaller granularity, reducing redundancy and unnecessary data invalidation. Dynamic address allocation better exploits parallelism within the SSD by assigning I/O operations to different planes based on current resource usage. Our findings suggest that optimizing the SSD configuration in this way for GPUDirectStorage systems can significantly speed up the processing of large applications, which can increase innovation turnaround in fields such as healthcare, finance, and language. Furthermore, the reduced data movement and enhanced resource utilization can result in longer SSD lifespans and less energy consumption, providing a more cost-effective solution to meet data processing demands.UndergraduateComputer Scienc
President's Newsletter
No full textOur Interdisciplinary Research Institutes bring together researchers from all seven Colleges and the Georgia Tech Research Institute to advance innovation and technology in sectors of strategic importance — think bioengineering, energy, nanoscience, renewable materials, and others — in concert with our government and industry partners. This summer, we launched two more: the Institute for Neuroscience, Neurotechnology, and Society (INNS) and the Space Research Institute (SRI). Both come at a critical time for our country and our world
Variational approaches to 3D reconstruction from multiple depth images
No full textThis dissertation introduces a novel variational framework for reconstructing 3D surfaces from depth data acquired by commercial sensors. The work is organized into three main contributions.
First, by leveraging variational methods and differential geometry, we derive explicit expressions for occluding boundaries—a key source of non-differentiability in 3D-2D matching tasks. Unlike prior approaches that rely on curvature measures or implicit representations, our work characterizes the local structure of occluding curves directly from the surface geometry. In particular, we demonstrate that the tangent of an occluding curve can be extracted from the surface’s second-order structure along the viewing direction, decomposing naturally into components corresponding to geodesic torsion and normal curvature. This result not only clarifies the role of occluding boundaries but also integrates seamlessly into our implicit reconstruction framework.
The second contribution addresses the challenges of reconstructing surfaces from raw, noisy, and incomplete depth maps—a common limitation of commercial depth cameras. We propose a variational framework that enforces global data fidelity while incorporating flexible regularization strategies. A key innovation is the separation of foreground and background modeling, which minimizes the influence of erroneous background data. To further mitigate issues from missing measurements, we introduce a novel inpainting technique that works in concert with an area-penalty regularizer. In addition, a new initialization scheme for the implicit function is presented, accelerating convergence and enhancing reconstruction accuracy, particularly for smooth surfaces.
Finally, we exploit the ideas from shape analysis, and develop a new regularizer for neural signed distance function (SDF) reconstruction. By directly addressing the instability introduced by the prevalent Eikonal loss, our regularizer not only improves convergence but also preserves finer geometric details compared to existing methods. Rigorous analysis and extensive experiments on public benchmarks validate its superior performance.
In summary, this dissertation demonstrates that integrating variational methods and shape analysis can overcome significant data-quality challenges in 3D reconstruction, while also offering new insights that enrich learning-based approaches.Ph.D.Electrical and Computer Engineerin
Signals of Change: How #MeToo Echoes in Computing Conferences
No full textThis study presents a bibliometric analysis of gender representation across sub-disciplines of computer science—an area historically overlooked despite persistent gender disparities. It compares the gender of authors and keynote speakers at computing conferences, with keynote speakers recognized as influential figures and markers of recognition within their fields. Findings indicate a general incline in gender parity within computer science academia, as measured through conference submissions and invited talks. From 2004 to 2024, the subfields software engineering and human-computer interaction consistently showed the highest levels of female representation. Notably, a rise in female keynote speakers was observed from 2016 to 2019, followed by a decline in 2024. This temporary increase may be linked to the broader cultural impact of the #MeToo movement beginning in 2017. A primary limitation of the study was the reliance on manual web scraping and the use of libraries with limited gender inference capabilities. Future research could further investigate the causal mechanisms behind these trends and explore more robust gender classification tools to strengthen findings.UndergraduateComputer Scienc
Integration of a Chromatic Aberration Technique for In Situ Working Distance Measurement During Powder Blown Laser Directed Energy Deposition Additive Manufacturing
No full textPowder blown laser Directed Energy Deposition (DED) is an additive manufacturing process that is increasingly being adopted in industry for its capability of realizing complex parts with several metal alloys. However, because of the nature of the process, the height of the deposited material can slightly deviate from the intended height, which affects the working distance (the distance between the deposition nozzle and the material deposited). This issue leads to inconsistency in deposited layer thickness and also to inferior mechanical properties. Such errors, if being accumulated, can generate significant discrepancies between the desired model and the printed object, resulting in a waste of time and material. Therefore, controlling the working distance is critical in optimizing the process.
To address this issue, this research develops a sensor that can in-situ measure the working distance and can ultimately be used for real-time control of the DED process. The sensor takes the radiation from the molten metal in the melt pool as a light source, which eliminates the needs for any additional space-taking light sources. The sensing mechanism is based on the optical phenomenon called the chromatic aberration – wavelength-dependent focal distance.
A lens is added to the existing optical system inside the printing head to collect lights radiating from the melt pool which are then delivered to a detecting optics through a fiber optic cable. The detecting optical system filter and process the lights in two wavelength bands. Changing the working distance changes the focal lengths of these two wavelength bands and accordingly their signal intensities. Therefore, these relative intensity changes are directly related to the changes in the working distance.
The sensor is calibrated for two different materials using a series of single-track prints; the heights of these single-tracks are then measured using a profilometer. Since the radiation spectrum depends on the temperature of the emitting surface; the calibration curve is unique to each material with its specific melting temperature. For this reason, two materials with very much different thermal and physical properties are tested. A steel alloy is chosen for its widespread industrial use, while an aluminum alloy is chosen to take advantage of the feature of the printer that allows to operate in a fully inert atmosphere. A procedure to process the acquired data and calibrate the sensor is proposed. The raw signal acquired is first cleaned from outliers that can negatively influence the measurement. Then, the resulting voltages (one for each wavelength band) are combined into a single-value error metric that directly correlates with a working distance. The results demonstrate that the developed sensor can precisely estimate the working distance and so be used as feedback control for the printing process. The sensor is completely passive, not requiring additional laser sources, and its sampling frequency is higher than those of existing camera-based techniques. Moreover, the sensor is designed as an add-on to the existing print head and does not interfere with the workspace of the DED system. Its compact design facilitates its applicability into powder-blown DED systems. Possible future work and optimization are also discussed.M.S.Mechanical Engineerin
Machine Learning Approaches for Knowledge Discovery in Nanophotonic Structures
No full textIn our modern societal context, Artificial Intelligence (AI) algorithms, particularly those rooted in machine learning (ML) and deep learning (DL), exert significant influence over our daily routines, owing largely to the convergence of abundant data accessibility and formidable computational capabilities. ML methodologies have evolved into indispensable tools across diverse domains, spanning science, engineering, medicine, and beyond, facilitating tasks such as material innovation and medical data analysis. Notably, ML algorithms have found particular utility in inverse design and knowledge extraction within nanostructures, surpassing conventional methods by virtue of their adeptness in processing high-dimensional datasets and revealing intricate data-structure relationships. However, a critical challenge arises from the opaque nature of these ML algorithms, which often function as enigmatic black boxes, obscuring the rationale behind their decision-making processes. This opacity poses a substantial barrier for engineers striving to optimize device precision. To address this challenge, this thesis undertakes a systematic exploration focused on elucidating the knowledge embedded within ML algorithmic decisions. This endeavor employs two distinct methodologies: firstly, pruning techniques streamline neural network architectures to unveil the precise impact of each input on the output, thereby enhancing interpretability and efficiency. Secondly, interpretable models, such as SHAP (SHapley Additive exPlanations), rooted in Game Theory, provide a comprehensive understanding of the contribution made by each design parameter to the decision outcome, enabling a comparative analysis of different approaches across practical metaphotonic structures. The thesis concludes with a thorough investigation aimed at identifying the most efficient neural network architecture for inverse design in nanophotonic structures, considering both statistical and computational complexities inherent in such design endeavors.Ph.D.Electrical and Computer Engineerin
Anti-Jamming Measures for Distributed Satellite Systems
No full textThis thesis presents advanced techniques to enhance the Anti-Jamming (AJ) capability of military multi-satellite communication systems, ensuring resilience against adversarial interference. The focus is on improving the Protected Tactical Communications system’s resilience, particularly during the initial acquisition phase, when users attempt to establish a connection for the first time. In addition to developing multi-satellite combiner techniques, this work explores enhancements in receiver processing blocks used by individual satellite systems during acquisition.
This thesis introduces a multi-satellite combiner based on the Signal-to-Noise Ratio (SNR) combiner to enhance Anti-Jamming (AJ) performance while accounting for variations in noise power across satellites. Its performance is compared with traditional diversity combiners, including the Minimum Mean Square Error (MMSE) combiner and the Maximal Ratio Combiner (MRC). The thesis evaluates the effectiveness of each method under various jamming models and Additive White Gaussian Noise (AWGN) conditions.
The results demonstrate that the proposed Anti-Jamming (AJ) framework, which includes the SNR-based multi-satellite combiner, significantly improves reception performance in the presence of jamming when integrated with enhanced Direction-of-Arrival
(DoA) estimation and frame synchronization algorithms. These enhancements, when incorporated into the existing Protected Tactical Waveform (PTW) standard, contribute to a more robust and resilient military satellite communication system.M.S.Electrical and Computer Engineerin
Competitive Binding of Synthetic Oligonucleotides
No full textThe work described in this dissertation focuses on competitive binding of synthetic oligonucleotides. Aptamers are single-stranded oligonucleotides capable of binding to non-nucleotide targets with high specificity and sensitivity, and can be used in a variety of applications including diagnostics, therapeutics, and molecular sensing. The traditional approach for aptamer screening known as SELEX (Systematic Evolution of Ligands by EXponential enrichment) requires PCR amplification of bound sequences following each incubation cycle; however, this introduces sequence bias and unwanted side-products. In order to mitigate these issues, the Milam lab developed a PCR-free alternative to SELEX known as CompELS (Competition-Enhanced Ligand Selection). While previous CompELS experiments have yielded aptamer candidates, the process remains a “black box” as sequences are only identified at the end of the screening.
Two separate CompELS screening experiments are discussed in this dissertation: one against a biological target (Fc protein) and against with a material target (graphene). The Fc protein is the conserved region of IgG antibodies, therefore Fc-binding aptamers could serve as “universal handles” for antibody capture and purification. Graphene is an allotrope of carbon, whose unique electronic properties are useful in sensing applications. Both of these CompELS experiments used “barcoded” screening libraries which could be used to determine which cycle of CompELS a particular sequence was added to the target. The results of these barcoded CompELS experiments indicated that there was indeed an exchange of bound sequences occurring at each successive cycle. Further discussion of the CompELS is presented in Chapter 2, Chapter 4, and Chapter 5.
Following CompELS—or any aptamer screening experiment—it is necessary that the aptamer candidates be sequenced. Incredible progress in DNA sequencing has been made in the past two decades; however, these techniques are largely optimized for the sequencing of genomic DNA rather than the random library sequences used in aptamer screening. While genomic DNA fragments vary in length and composition, aptamer screening library sequences do not: each sequence consists of a random segment flanked by fixed base segments. This uniformity in base composition of aptamer screening libraries results in a dramatically decreased sequencing yield. The work presented in Chapter 3 provides insight into the causes of these sequencing challenges, as well as suggestions for mitigating these issues.
In addition to the aptamer research discussed above, this dissertation also presents a platform for the detection of specific nucleic acid sequences using synthetic oligonucleotides. This project began during the early months of the Covid-19 pandemic, when the reagents and supplies for qPCR-based testing methods were in short supply. This goal of this project was to develop a method a PCR-free method of detecting viral RNA sequences using flow-cytometry. This detection platform relied on double-stranded oligonucleotide probe systems: one strand (complementary to the target nucleic acid sequence of interest) was conjugated to the surface of a microsphere and possessed a fluorescent tag, while its hybridization partner possessed a quencher molecule. The two strands comprising a dsprobe system must be chosen such that the duplex remains stable under normal conditions, while still allowing for displacement of the quencher-capped strand by the target nucleic acid sequence. When the fluorescently-tagged probe sequence and its quencher-capped hybridization partner are hybridized, the system has a low fluorescence signal; however, when the target nucleic acid sequence displaces the quencher-capped strand, the system has a high fluorescence signal.
Chapter 6 describes the various probe systems that were screened for sensitivity and specificity, resulting in the selection of a particular probe system for further characterization. Titration studies revealed that while the selected probe system was highly responsive to the target nucleic acid sequence, the fluorescence response of the system was unexpectedly low. The work presented in Chapter 7 revealed that the quencher molecules used in this study (Iowa Black quenchers) exhibited significantly longer range quenching effects than anticipated. As a result of this long-range quenching effect, even slightly incomplete displacement of the quencher-capped strands resulted in significantly decreased fluorescence signal.Ph.D.Materials Science and Engineerin