DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Glitch serendipity: exploring the spectrum across media
No full textAugust 2023School of Humanities, Arts, and Social SciencesThis thesis presents glitch serendipity as a formalized approach to artistic practice. The study focuses on the concepts of glitch, remix, and serendipity, employing interdisciplinary methods and principles such as disjunctive strategies, explorability, and ambiguity. Glitch serendipity offers researchers flexible methods for open exploration, leading to unexpected encounters and generating research prospects. This approach enhances our understanding of phenomena in an interdisciplinary manner, facilitating a breadth of experiences and expressions. Through explorative methods, remixing of digital materials, and leveraging open-source tools, novel ways of experiencing and interacting with media are attained. The portfolio reflects on projects across media, including still imagery, gaming, video art, and live coding to create a spectrum of glitch expression, highlighting the unique insights and possibilities they offer. The formalized approach outlined in this thesis pushes the limits of digital structures, encourages experimentation with conceptual juxtapositions, and embraces ambiguity, resulting in serendipitous insights and discoveries that open new avenues for artistic expression and challenge normative perceptions of functionality and beauty.MF
Numerical investigation of k<sub>σ</sub> at high confining pressures under field drainage conditions
No full textMay 2023School of EngineeringThe State of Practice (SoP) for assessing liquefaction triggering of cohesionless soils at high confining pressures extrapolates the SoP liquefaction triggering charts using the overburden correction factor (Kσ). The SoP liquefaction triggering charts are calibrated using earthquake case histories at relatively shallow elevations with vertical effective stresses less than 2 atm, while the SoP overburden correction factor relationships are either based on undrained cyclic laboratory tests or low-confining pressure liquefaction field case histories (σ values greater than 1.0, much greater than those from the SoP relationships. With these experiments being representative of some possible field conditions at high confining pressures, further research is needed to explore the behavior of Kσ at high confining pressure under different scenarios, which can be practically achieved using numerical simulations and parametric studies. The work presented herein calibrates the numerical program FLAC and the constitutive model PM4Sand to the 45-1 and 45-6 single-drainage centrifuge experiments by Dr. Min Ni at the system level, and extends these experimental results via a parametric study to study the effect of varying the sand layer permeability and thickness on the pore water pressureresponse and Kσ. The results from the numerical calibration are found to be generally in very good agreement with the centrifuge experiments at both confining pressures, with an excellent agreement in the excess pore water pressure ratio and the shear stress time histories. In addition, the Kσ from the numerical model is found to be 1.29, which is greater than 1 and very similar to the 1.28 value from the centrifuge experiments. The calibrated models are then prepared for a three-part parametric study. Part I studied the effect of varying the Ottawa sand permeability on the soil response when exposed to the same earthquake. This revealed the need of not restricting the location where Kσ is calculated to the bottom of the soil, since the elevation where the maximum excess pore water pressure ratio reaches 0.8 changes as the permeability changes. This led to re-evaluation of the calibrated runs to determine the exact rumax = 0.8 location, with corresponding recomputation of Kσ, which increased from about 1.3 to 1.51. Part II of the parametric study evaluated the effect of varying the permeability of the Ottawa sand layer on Kσ with scaling up or down of the input motion in each case to reach a target maximum excess pore water pressure ratio of 0.8 in the whole sand layer. The study showed an undrained Kσ greater than 1.0 (= 1.27), in contradiction with the cyclic undrained triaxial results on Ottawa sand from Dr. Min Ni’s dissertation and the cyclic undrained results on sands from the literature. However, the study provided an increasing Kσ versus permeability trend that further supports the results from the centrifuge tests by Dr. Min Ni. Kσ is found to increase with permeability until an approximate permeability of 0.001 cm/sec, where Kσ levels off at around 1.50. Additionally, the location where the maximum excess pore water pressure ratio occurs is found to move downwards in the soil profiles as the permeability increases, for both confining pressures of 1 and 6 atm. Part III of the parametric study focused on the effect of varying the thickness of the Ottawa sand layer on Kσ. It revealed that Kσ decreases as the soil thickness increases, down to an approximate thickness of 5 m, and then levels off at an approximate value of Kσ = 1.50, which is the same value at which Kσ leveled off for a thickness of 5 m when increasing the permeability from zero to 0.012 cm/sec. Moreover, the location where the maximum excess pore water pressure ratio occurs is found to move upward in the soil profile as the soil thickness increases, regardless of confining pressure, which is the opposite to the behavior noticed when the soil permeability increases.Ph
TG-CSR: A human-labeled dataset grounded in nine formal commonsense categories
No full textMachine Common Sense Reasoning is the subfield of Artificial Intelligence that aims to enable machines to behave or make decisions similarly to humans in everyday and ordinary situations. To measure progress, benchmarks in the form of question-answering datasets have been developed and published in the community to evaluate machine commonsense models, including large language models. We describe the individual label data produced by six human annotators originally used in computing ground truth for the Theoretically-Grounded Commonsense Reasoning (TG-CSR) benchmark's composing datasets. According to a set of instructions, annotators were provided with spreadsheets containing the original TG-CSR prompts and asked to insert labels in specific spreadsheet cells during annotation sessions. TG-CSR data is organized in JSON files, individual raw label data in a spreadsheet file, and individual normalized label data in JSONL files. The release of individual labels can enable the analysis of the labeling process itself, including studies of noise and consistency across annotators
Optoelectronic properties of layered perovskites with strong spin-orbit coupling
No full textAugust 2022School of EngineeringSpintronic devices, by harnessing the spin degree of freedom, are expected to outperform charge-based devices in terms of energy efficiency and operation speed. For technological relevance, the use of electric field to control the spin degree of freedom at room temperature has been pursued for several decades. A major hurdle that leads to slow progress is the dilemma between effective control and strong spin decoherence. For example, in a Rashba or Dresselhaus material with strong spin-orbit coupling, though the internal magnetic field could be large enough for effectively controlling the spin precession, the spin dephasing time in most cases inevitably becomes extremely short through Dyakonov-Perel scattering. To address such a dilemma, persistent spin helix – a spin density wave with its phase and amplitude immune to spin-independent scattering – has been proposed in systems with SU(2) symmetry. However, most materials systems by far hosting persistent spin helix are carefully engineered III-V quantum wells that operate at cryogenic temperature with unwanted long spin helix wavelength. In this dissertation, we show the discovery of persistent spin helix in an organic-inorganic hybrid ferroelectric halide perovskite (4,4-DFPD)2PbI4 (4,4-DFPD: 4,4-difluoropiperidinium) whose layered nature makes it intrinsically like a quantum well. We demonstrate that the spin-polarized band structure is switchable at room temperature via intrinsic ferroelectric field. We reveal the valley-spin coupling through circular photo-galvanic effect in single crystalline bulk crystals. The favored short spin helix wavelength (three orders of magnitude shorter than III-V), room temperature operation and nonvolatility make the hybrid perovskite an ideal platform in understanding symmetry-tuned spin dynamics towards designing practical spintronic materials and devices that can resolve the control-relaxation dilemma.
For controllable growth and processing for miniaturized spintronic devices, we further show the liquid-phase van der Waals epitaxy of (4,4-DFPD)2PbI4 on muscovite mica and demonstrate the feasibility on fabricating perovskite-perovskite vertical heterostructures using dissimilar Riddlesden-Popper 2D perovskite sheets. The epitaxial (4,4-DFPD)2PbI4 nanobelt array can be from multiple layers to unit-cell in thickness and are crystallographically aligned on the mica substrate. An interlayer photo emission in (4,4-DFPD)2PbI4-based heterostructure with a lifetime of about 25 ns at 120 K has been revealed. Our demonstration of epitaxial (4,4-DFPD)2PbI4 array grown on mica via liquid-phase van der Waals epitaxy provides a paradigm to prepare orderly distributed 2D hybrid perovskites for further integration into multiple heterostructures. The discovery of a new interlayer emission in (4,4-DFPD)2PbI4-based heterostructure enriches the basic understanding of interlayer charge transition in halide perovskites systems.Ph
High-fidelity information extraction in future power grids
No full textDecember 2022School of EngineeringThe recent decades have witnessed the rapid growth of the deployment of smart meters in power systems. The massive amount of data collected from smart meters provide rich information of power systems, such as system dynamics and load profiles. The reliable, high-fidelity data processing is beyond the capability of the existing data analytic tools of power system monitoring. The goal of this research aims to bridge the gap between power system monitoring and high dimensional data analysis. One of the key observations is that many data in power systems have intrinsically low dimensionality, although the original data are high dimensional. This dissertation focuses on high-fidelity information extraction by exploiting these low-dimensional structures. The first part of this dissertation studies one low-rank model, dictionary learning, with applications in energy disaggregation at the substation level (EDS). The high penetrations of renewable distributed energy resources, such as solar generations, are usually behind-the-meter (BTM) and thus are invisible to the power system operators. The uncertainty introduced by the BTM renewable generations brings great challenges to reliable power system planning and operation. This dissertation formulates the energy disaggregation tasks as a dictionary learning problem and obtains accurate estimations of load profiles from aggregate measurements. Because it is difficult to obtain full labels for training data, this dissertation, for the first time, addresses the “partial labels” issue in EDS and proposes a dictionary learning-based EDS method to disaggregate each load from aggregate measurements in real-time. In the offline training stage, a novel column-sparsity constraint is added by exploiting the group sparsity of unlabeled loads, and an incoherence regularization term is proposed to promote discriminative patterns for each load. In the online disaggregation stage, the traditional sparse decomposition approach is improved by decomposing the aggregate measurements as a linear combination of some representative learned disaggregations. Because of the invisibility and stochastic nature of BTM renewable generation, it is inevitable that the disaggregation results contain errors. However, existing EDS approaches cannot quantify the uncertainty of the disaggregation results. This dissertation then studies the EDS problem from a Bayesian perspective and models the EDS as a Bayesian dictionary learning problem. A scheme to measure the uncertainty of the disaggregation results is proposed. In the offline training stage, the proposed approach learns the probabilistic distributions of dictionaries and coefficients from the aggregate training data with partial labels. In the online disaggregation stage, the proposed approach computes the predictive mean and covariance of the probabilistic distribution of each load consumption. The mean is used as the load estimation andthe covariance is employed to provide the uncertainty measure of the disaggregation results. The second part of this dissertation studies another low-rank model, the robust matrix completion, with applications in synchrophasor data recovery. Phasor measurement units (PMUs) offer high-resolution synchrophasor measurements and thus provide better visibility of the power system dynamics. However, the synchrophasor data quality issues, such as missing data and bad data attacks, hinder the PMU data from being incorporated into the power system operation and control and prevent the large-scale deployment of PMU in North America.This dissertation formulates the synchrophasor data recovery problem as a Bayesian robust Hankel matrix completion problem. In particular, the proposed method exploits the low-rank Hankel property of synchrophasor data to recover the consecutive and simultaneous data loss/corruption, where the standard low-rank matrix completion methods cannot handle this extreme case. The proposed Bayesian method learns the probabilistic distributions of decomposed factors and infers the distribution of synchrophasor data. Compared with existing data recovery methods, this dissertation, for the first time, provides the uncertainty measure of the returned results. The low-rank Hankel approximation of the synchrophasor data matrix assumes that the data are generated from a linear dynamical system. When a significant event takes place in power systems, the nonlinear synchrophasor data do not hold the low-rank property. This dissertation then studies the lifted low-rank Hankel property of nonlinear synchrophasor data in a higher dimensional space by exploiting the kernel trick. The idea of the proposed nonlinear synchrophasor data recovery method is to lift the Hankel matrix of nonlinear synchrophasor data into a higher dimension such that the lifted Hankel matrix is low-rank in that high dimensional space. The kernel trick is employed to perform the implicit lifting. The proposed Bayesian framework can also provide an uncertainty index to measure the uncertainty level of the recovery results.Ph
RPIrates: Fun with OpenAI, GPTStudio and R!
No full textSeemingly everyone has been talking about the impact of OpenAI's ChatGPT on, well, everything, including writing code. In this very special RPIrates we talk specifically about generating great, and sometimes not-so-great, R code based on OpenAI's "Codex" models, which are easily accessed via the OpenAI API with the help of RStudio extensions (addins) provided by the GPTStudio package. After a quick icebreaker in which we demo GPTStudio in action, we briefly review transformer neural networks; we talk about how code completion tools like Microsoft's IntelliCode Compose apply TNN frameworks to write scarily excellent code; we tour OpenAI's Codex documentation; and then we get back to more hands-on with GPTstudio. We also talk about why code completion frameworks like Codex are so great at Python and Javascript but spotty with R, and of course, the ethics; oh, the ethics! Many links for further learning and research are provided
Synthetic data generation and evaluation for fairness
No full textAugust 2023School of ScienceArtificial Intelligence (AI) models often have unfairness, resulting in biased predictions against certain groups of protected individuals. This thesis addresses two broad objectives for quantifying fairness in AI. As the first objective, we define the novel problem of unfairness at the subgroup-level in the context of privacy-preserving synthetic data, especially healthcare data. This is followed by the introduction of time-series and disparate impact fairness metrics for measuring the resemblance (similarity) between the real and synthetic data. For the second objective, we audit the fairness of Machine Learning models and bias mitigation algorithms by stress-testing them under shifts. The thesis describes two auditing pipelines: (a) Fairness Auditor: Grid-based auditing using Iterative Proportional Fitting and (b) Adversarial Auditor: Adversarially attacking utility and fairness objectives using Multi-Objective Bayesian Optimization. Although healthcare data is abundant, access to it is often restricted by privacy laws, and thus, synthetic data provides a viable alternative. Current research measures fairness in various forms but does not discuss the problems of unfairness in synthetic data. Thus, we address the novel problem of defining and then, quantifying the fairness of synthetic data, considering both temporal and non-temporal datasets. Here, we address two definitions of fairness: (a) Machine Learning (ML) fairness of synthetic data and (b) Representational bias in synthetic data. The results highlight that synthetic data exhibits variable bias properties from the real data, as measured by both group fairness metrics on trained ML models and fairness metrics for subgroup-level resemblance. With Machine Learning models and bias mitigation algorithms being used for real-world tasks where shifts in data are common, their applicability under such shifts without robust testing is unknown. This hinders trust in AI models and raises concerns about their utility and fairness under shifts. Here, we discuss how conditional sampling of synthetic data can be used for robust stress-testing of Machine Learning models and bias mitigation algorithms under shifts to identify fairness vulnerabilities (change in biases). Each method is robustly evaluated using the auditing pipelines and summarized using Fairness Reports and novel metrics. These auditing pipelines are a step towards ensuring that Machine Learning methods can be applied under shifts, building trust in these algorithms before real-world application. The results demonstrate that these methods have variable utility and fairness scores on different datasets and lead to increased biases under certain shifts. The thesis concludes with the contributions and discusses potential future work on how the insights can be used to extend Machine Learning auditing to non-binary protected attributes and other tasks such as regression and clustering.Ph
Preprocessing and learning for graph structured data
No full textAugust 2023School of ScienceGraphs are general structures which may be used to describe any system or dataset with related elements. Because of the prevalence of such data, efficient and accurate algorithms for analyzing graphs are of extreme importance in numerical algorithms and general data science tasks. The size of graph structured datasets has only increased in the past decades and promises to continue doing so. As companies like Google and Meta wrestle with peta and exa-scale graph analysis problems; computational scientists face many of the same issues as simulations require ever-larger meshes. Because of this, acceleration and preprocessing techniques are important to ensure graph algorithms run efficiently and accurately. We investigate several preprocessing and acceleration techniques for performing tasks on graph structured data. We develop a methodology for generating graph null-models with a desired degree distribution. This a problem which has been of interest to network scientists for decades. Despite this, parallelizable, fast subroutines used in algorithms for generating such graphs tend to yield inaccurate distributions. We suggest a novel analysis technique for the popular Chung-Lu random graph generator, and show that this analysis technique provides a method for automatically generating parameters for Chung-Lu-like null models as a pre-processing step. We provide several methods for generating these null-models, and show that in all cases we significantly out-perform standard Chung-Lu generation. We also suggest that such null models may be used to improve the accuracy of Modularity maximization. Additionally, we examine the task of coarsening graphs while preserving the spectrum of the graph Laplacian. Coarsening is an important preprocessing step for many large scale graph problems which aim to solve relatively smal subproblems and reconstruct an approximate solution on the original graph. Coarsening is used in clustering, partitioning, and multigrid methods for solving linear systems of equation. The graph Laplacian is an important operator for describing graph structured data. It relates the heat transfer in a graph to its topology. As such, its eigenvectors and eigenvalues hold important information about edge cuts and clustering. We present a heuristic for preserving the spectrum of the graph Laplacian during coarsening, and present a parallel algorithm for utilising this heuristic. This is in contrast to prior publications on the subject which focus on serial and k-means methods for spectrum consistent coarsening. We further analyze the inverse problem, and find that the original graph may be reconstructed to within some edge-weight error given a coarse representation which approximates its spectrum. This presents a novel development in graph coarsening literature and suggests that preserving a graphs spectrum during coarsening may be sufficient to preserve all structure. Finally we investigate a technique for accelerating the training of graph neural networks using Koopman operator theory. Graph neural networks provide a powerful method for performing classification and prediction tasks on graphs. This is in contrast to traditional neural networks which struggle with the unordered nature of nodes and edges. Because of this, a great deal of effort has been put into accelerating graph neural networks through techniques such as graph pooling. Despite this, they are still often slow to train. We suggest a method for accelerating training by interweaving standard backpropagation steps with prediction steps that make use of simple matrix-vector multiplication. We apply our method to the task of node classification and find that it is prone to instability, but can achieve multiple times speed-ups over Adam for well-chosen parameters. This work represents the first time Koopman training has been applied to graph neural networks, and the first time it has been applied on GPU.Ph
Stochastic hydrodynamics of colonial microswimmers
No full textMay 2021School of ScienceFunctions of microorganisms like looking for nutrition and biological processes such as infectionand reproduction are only possible due to micro-fluidic motion. Developments over the
past two decades in tracking and manipulation at the micro-fluidic level have made possible
accurate measurements of micro-hydrodynamical flows and have provided access to a wealth
of interesting biological and synthetic phenomena and a means to test the phenomena's
corresponding theories. This thesis focuses on these corresponding theories of micro-hydrodynamical phenomena.Our aim is to answer questions of individual and collective mobility for colonial microswimmers
and micro-rotors and of the fluid flows they generate. We wish to give analytical
expression to the individual aggregated swimmers' mobility statistics, and the environmental
interaction of these swimmers both with each other and with reactive proles such as
attractant chemicals. Our work is effectively split into three projects. The first research project focuses onthe statistical mobility properties of colonial microswimmers typified by an animal-like group
of protozoa called Choanoflagellates. The second research project focuses on the function of
multicellularity for the taxis and kinesis effectiveness of protozoa. Finally, the third research
project focuses on the stability and correlations of suspensions of micro-swimmers and microrotors.
In our investigations we applied fluid mechanical principles for the flow modeling and
insights from biophysical experiments for the swimmer modeling. We used asymptotic and
multi-scale methods for coarse graining the physical models and we compared the results from
these asymptotic derivations with Monte Carlo simulations and matrix-based computations
of the full models.Ph
Linking mineral reactivity to its semiconducting band structure
No full textMay 2023School of EngineeringCertain transition metal oxides are known to readily reduce chromium ion from its highly toxic +VI oxidation state to its +III state, which is an important nutrient for human biological functions. Herein, the factors which govern the reduction of Cr (VI) to Cr (III) by five selected Fe and Mn based metal oxides were investigated in order to better understand the mechanism driving the reduction. Results show that of the five metal oxides (hematite, hollandite, magnetite, manganite, and Romanechite), only Romanechite displayed the capability of reducing the concentration of Cr (VI) in isolation. This distinctive capability is likely related to their differences in their electronic properties, namely the differing band alignment with respect to Cr (VI) / Cr (III) redox couple in the aqueous environment.M