Selectively Decentralized Reinforcement Learning

Nguyen, Thanh Minh Purdue University ProQuest Dissertations & Theses 2018 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The main contributions in this thesis include the selectively decentralized method in solving multi-agent reinforcement learning problems and the discretized Markov-decision-process (MDP) algorithm to compute the sub-optimal learning policy in completely unknown learning and control problems. These contributions tackle several challenges in multi-agent reinforcement learning: the unknown and dynamic nature of the learning environment, the difficulty in computing the closed-form solution of the learning problem, the slow learning performance in large-scale systems, and the questions of how/when/to whom the learning agents should communicate among themselves. Through this thesis, the selectively decentralized method, which evaluates all of the possible communicative strategies, not only increases the learning speed, achieves better learning goals but also could learn the communicative policy for each learning agent. Compared to the other state-of-the-art approaches, this thesis's contributions offer two advantages. First, the selectively decentralized method could incorporate a wide range of well-known algorithms, including the discretized MDP, in single-agent reinforcement learning; meanwhile, the state-of-the-art approaches usually could be applied for one class of algorithms. Second, the discretized MDP algorithm could compute the sub-optimal learning policy when the environment is described in general nonlinear format; meanwhile, the other state-of-the-art approaches often assume that the environment is in limited format, particularly in feedback-linearization form. This thesis also discusses several alternative approaches for multi-agent learning, including Multidisciplinary Optimization. In addition, this thesis shows how the selectively decentralized method could successfully solve several real-worlds problems, particularly in mechanical and biological systems.

Effects of Calcium Hydroxide Treated Corn Stover Inclusion as a Partial Forage Replacement in Diets for Lactating Dairy Cows

Casperson, Brittany Purdue University ProQuest Dissertations & Theses 2017 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

The central hypothesis of this dissertation is that calcium hydroxide treatment would improve the feeding value of corn stover and treated corn stover could replace a portion of the traditional forages fed to lactating dairy cattle. There are four main objectives addressed in this dissertation to test the central hypothesis. First, evaluate the effects of pretreatment of corn stover with calcium hydroxide. Second, determine the effects of replacing alfalfa haylage in the diet of dairy cows with calcium hydroxide-treated corn stover and the effect of further replacement of corn silage on dry matter intake, milk production, and milk composition. Third, determine the effects of maximal replacement of either alfalfa haylage or corn silage with treated corn stover on feed intake, total tract digestibility, rumen fermentation parameters, rumen digestibility, and milk production in order to assess the origins of increased feed efficiency of mid-lactation dairy cows fed alkaline treated corn stover, and lastly, to evaluate the effects of feeding a fortified corn stover pellet, pelleted treated corn stover (PTCS), on dry matter intake, milk production, milk composition, total tract digestibility, rumen fermentation parameters, and rumen digestibility. Furthermore, the effect of physical form of the PTCS ingredients were evaluated by comparing PTCS ingredients fed in pellet or non-pellet form. Calcium hydroxide treated corn stover was first evaluated in a lactation performance trial where treated corn stover was fed in a TMR at 0% of the diet DM or replaced either alfalfa haylage or alfalfa haylage and a portion of the corn silage at 15 or 30% of the diet DM, respectively. After 10 weeks of feeding, cows fed treated corn stover to replace alfalfa haylage resulted in reduction (P < 0.05) of DMI. Subsequently, milk production, milk composition, and energy corrected milk production were not different (P > 0.05) between the treatments. However, energy corrected milk per unit of DMI (kg/kg), a measure of feed efficiency, tended ( P < 0.10) to be improved. Cows fed treated corn stover had consistent DMI over the 10-week feeding period compared to cows fed no treated corn stover which had slight increases in feed intake, over time. To determine the overall impact on feed efficiency with the inclusion of treated corn stover, we evaluated inclusion of treated corn stover to maximally replace either alfalfa haylage at 15% of the diet DM or corn silage at 19% of the diet DM. When treated corn stover was included in the diet to replace either traditional forage, we saw a reduction (P 0.05) in milk production, confirming the results of the previous study. Furthermore, percentage of milk fat was reduced (P < 0.05) with the inclusion of treated corn stover but there was no difference in yield of milk fat (P > 0.05). Energy corrected milk production per unit of DMI (kg/kg) was greater for cows fed diets containing treated corn stover. Inclusion of treated corn stover resulted in decreased (P < 0.05) apparent total tract digestibility of DM, OM, and CP when replacing corn silage. There was no difference (P > 0.05) in NDF digestibility, suggesting NDF digestibility was successfully improved with calcium hydroxide pretreatment. There was no difference (P > 0.05) on rumen pH or VFA concentrations. Improved feed efficiency is supported by changes in diet digestibility and the potential for changes in post-ruminal digestion. Improvements in feed efficiency using treated corn stover are promising but implementation of such feeding strategies may be limited to certain regions near site of harvest as the bulk density of corn stover is low. Therefore, to overcome this limitation for implementation into diets for lactating cows, we evaluated feeding treated corn stover as part of a fortified pellet- pelleted treated corn stover (PTCS). The PTCS pellet was included at 21 or 40% of the diet DM. The ingredients of PTCS were fed individually in a third treatment to match the 40% PTCS and evaluate if physical form had an impact. Dry matter intake was reduced (P < 0.05) with higher inclusion levels (40%) but physical form had no impact on DMI. Milk production and 4% energy corrected milk did not differ (P > 0.05) among the dietary treatments. Milk fat percentage was reduced (P < 0.05) with the inclusion of PTCS. Digestibility of DM and OM were reduced ( P < 0.05) with inclusion of treated corn stover. The inclusion of PTCS resulted in decreased (P < 0.05) acetate and increased (P < 0.05) propionate concentration, yielding a reduced (P 0.05) between dietary treatments. (Abstract shortened by ProQuest.).

Enriched Isogeometric Analysis for Parametric Domain Decomposition and Fracture Analysis

Chen, Chun-Pei Purdue University ProQuest Dissertations & Theses 2020 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

As physical testing does not always yield insight into the mechanistic cause of failures, computational modeling is often used to develop an understanding of the goodness of a design and to shorten the product development time. One common, and widely used analysis technique is the Finite Element Method. A significant difficulty with the finite element method is the effort required to generate an analysis-ble mesh due to the difference in the mathematical representation of geometry CAD and CAE systems. CAD systems commonly use Non-Uniform Rational B-Splines (NURBS) while the CAE tools rely on the finite element mesh. Efforts to unify CAD and CAE by carrying out analysis directly using NURBS models termed Isogeometric Analysis reduces the gap between CAD and CAE phases of product development. However, several challenges still remain in the field of isogeometric analysis. A critical challenge relates to the output of commercial CAD systems. B-rep CAD models generated by commercial CAD systems contain uncoupled NURBS patches and are therefore not suitable for analysis directly. Existing literature is largely missing methods to smoothly couple NURBS patches. This is the first topic of research in this thesis. Fracture-caused failures are a critical concern for the reliability of engineered structures in general and semiconductor chips in particular. The back-end of the line structures in modern semiconductor chips contain multi-material junctions that are sites of singular stress, and locations where cracks originate during fabrication or testing. Techniques to accurately model the singular stress fields at interfacial corners are relatively limited. This is the second topic addressed in this thesis. Thus, the overall objective of this dissertation is to develop an isogeometric framework for parametric domain decomposition and analysis of singular stresses using enriched isogeometric analysis.Geometrically speaking, multi-material junctions, sub-domain interfaces and crack surfaces are lower-dimensional features relative to the two- or three-dimensional domain. The enriched isogeometric analysis described in this research builds enriching approximations directly on the lower-dimensional geometric features that then couple sub-domains or describe cracks. Since the interface or crack geometry is explicitly represented, it is easy to apply boundary conditions in a strong sense and to directly calculate geometric quantities such as normals or curvatures at any point on the geometry. These advantages contrast against those of implicit geometry methods including level set or phase-field methods. In the enriched isogeometric analysis, the base approximations in the domain/subdomains are enriched by the interfacial fields constructed as a function of distance from the interfaces. To circumvent the challenges of measuring distance and point of influence from the interface using iterative operations, algebraic level sets and algebraic point projection are utilized. The developed techniques are implemented as a program in the MATLAB environment named as Hierarchical Design and Analysis Code. The code is carefully designed to ensure simplicity and maintainability, to facilitate geometry creation, pre-processing, analysis and post-processing with optimal efficiency.To couple NURBS patches, a parametric stitching strategy that assures arbitrary smoothness across subdomains with non-matching discretization is developed. The key concept used to accomplish the coupling is the insertion of a "parametric stitching" or p-stitching interface between the incompatible patches. In the present work, NURBS is chosen for discretizing the parametric subdomains. The developed procedure though is valid for other representations of subdomains whose basis functions obey partition of unity. The proposed method is validated through patch tests from which near-optimal rate of convergence is demonstrated. Several two- and three-dimensional elastostatic as well as heat conduction numerical examples are presented.

Interactive Mitigation of Biases in Machine Learning Models

Busum, Kelly Van Purdue University ProQuest Dissertations & Theses 2024 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Bias and fairness issues in artificial intelligence algorithms are major concerns as people do not want to use AI software they cannot trust. This work uses college admissions data as a case study to develop methodology to define and detect bias, and then introduces a new method for interactive bias mitigation.Admissions data spanning six years was used to create machine learning-based predictive models to determine whether a given student would be directly admitted into the School of Science under various scenarios at a large urban research university. During this time, submission of standardized test scores as part of a student's application became optional which led to interesting questions about the impact of standardized test scores on admission decisions. We developed and analyzed predictive models to understand which variables are important in admissions decisions, and how the decision to exclude test scores affects the demographics of the students who are admitted.Then, using a variety of bias and fairness metrics, we analyzed these predictive models to detect biases the models may carry with respect to three variables chosen to represent sensitive populations: gender, race, and whether a student was the first in his/her family to attend college. We found that high accuracy rates can mask underlying algorithmic bias towards these sensitive groups.Finally, we describe our method for bias mitigation which uses a combination of machine learning and user interaction. Because bias is intrinsically a subjective and context-dependent matter, it requires human input and feedback. Our approach allows the user to iteratively and incrementally adjust bias and fairness metrics to change the training dataset for an AI model to make the model more fair. This interactive bias mitigation approach was then used to successfully decrease the biases in three AI models in the context of undergraduate student admissions.

Understanding the Formation of Distant Galaxies in the Context of Large-Scale Structure

Huang, Yun Purdue University ProQuest Dissertations & Theses 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Understanding the formation and evolution of galaxies is one of the most fundamental questions in modern astronomy. While it is widely accepted that galaxy formation needs to be understood in the context of cosmic structure formation of dark matter, a complex interplay of different physical processes that drive galaxy formation makes it challenging to elucidate how the large-scale environment of dark matter influences galaxies, particularly in their formative epoch (z & 2)As the most luminous nebular emission arising from star formation, Lyα provides a promising and effective tool to study the young universe and nascent galaxies. At z > 2, Lyα emission is redshifted into the visible window that is detectable by ground-based telescopes. Existing studies also suggest that strong Lyα-emitting galaxies represent a young and lowmass galaxy population and therefore are the best visible tracers of the large-scale structure of the distant universe.In this thesis, I present two complementary studies designed to address these questions using Lyα emission as a cosmological tool. In Chapter 2, I investigate the kinematics and spatial distribution of the gas-phase interstellar and circumgalactic media using compact and diffuse Lyα emission in and around distant galaxies. I also carry out a comprehensive characterization of how Lyα properties correlate with other galaxy properties and the environment that galaxies reside in. In Chapter 3, I explore how Lyα-emitting galaxies trace the large-scale structure characterized by other means; I also conduct a detailed investigation of the distribution of different 'types' of galaxies and H i gas in and around the most massive cosmic structure known to date. These investigations are informative in building clear expectations for the ongoing and upcoming experiments - including the Legacy Survey for Space and Time, James Webb Space Telescope, Dark Energy Spectroscopic Instruments, and Hobby-Eberly Telescope Dark Energy eXperiment - in obtaining a detailed picture of galaxy evolution in the context of their environments.

Instability Measurements on Two Cone-Cylinder-Flares at Mach 6

Benitez, Elizabeth K Purdue University ProQuest Dissertations & Theses 2021 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

This research focuses on measurements of a convective shear-layer instability seen naturally in quiet hypersonic flow. Experiments were carried out in the Boeing/AFOSR Mach 6 Quiet Tunnel (BAM6QT) at Purdue University. The BAM6QT provides low-disturbance hypersonic flow with freestream noise levels similar to what would be experienced by a flight vehicle. To obtain high-speed, off-the-surface measurements of the instability, a modified focused laser differential interferometer (FLDI) was first designed to work with the contoured Plexiglas windows available in the tunnel.A cone-cylinder-flare geometry was then selected to study the instabilities related to an axisymmetric separation bubble at Mach 6. The sharp cone had a 5◦ half-angle, while flare angles of 10◦ and 3.5◦ were tested to compare axisymmetric compression with and without separation, respectively. Under quiet flow, laminar separation and reattachment was confirmed by schlieren and surface pressure-fluctuation measurements. Coherent traveling waves were observed. These were attributed to both the second-mode instability, as well as a shear-generated instability from the separation bubble. The symmetry of the bubble was found to be highly sensitive to angle of attack. Additionally, by introducing controlled disturbances on the cone upstream of the separation, larger-amplitude shear-generated waves were measured while the second-mode amplitudes remained unchanged. Therefore, the sheargenerated waves were amplified moving through the shear layer, while the second mode remained neutrally stable. These appear to be the first measurements of traveling waves that are generated in the shear layer of a separation bubble in hypersonic flow.

Mass Spectrometric Methods Development for Identification of Drug/Herbicide Substances and Mutagenic Impurities, and Gas-Phase Reactivity Study of Phenylcarbyne Anions

Feng, Erlu Purdue University ProQuest Dissertations & Theses 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Mass spectrometry (MS) is a versatile analytical tool that is especially useful for identification of unknown compounds in mixtures when coupled with chromatography. In MS experiments, the analytes are ionized, separated based on their mass-to-charge (m/z) ratios, and detected. The molecular weight of the analyte can often be derived from the mass spectrum if stable molecular ions (M•+) or stable protonated/deprotonated analyte molecules ([M+H]+ or [M-H]- ) are generated. Further, MS can also be used to obtain structural information for the ionized analytes via their fragmentation reactions. Tandem mass spectrometry ( MSn ) experiments are powerful for the characterization of unknown compounds in mixtures without the need for coupling them with chromatography. In MSn experiments, the analytes are ionized, the ions of interest are isolated and subjected to reactions, such as collision-activated dissociation (CAD) or ion-molecule reactions with neutral reagent molecules. The fragmentation pattern or the diagnostic ion-molecule reaction product ions can be utilized to elucidate the structures of the analytes. The fragment ions or diagnostic product ions can further be subjected to CAD to obtain more structural information. Besides analytical purposes, MSnalso provides a powerful tool for exploring the reactivities of reaction intermediates that are elusive, such as phenylcarbyne anions and phenylcarbene anions.The research described in this dissertation mainly focuses on the development of MSnmethods based on diagnostic gas-phase ion-molecule reactions followed by CAD for (1) the characterization of differently substituted ureas and (2) the differentiation of sulfonate esters from their isomeric analogs, such as sulfite esters and sulfones. HPLC was coupled with the MSnmethods discussed above to demonstrate its usefulness in the identification of compounds in mixtures. Additionally, a gas-phase reactivity study on phenylcarbyne anions is discussed in this dissertation. The phenylcarbyne anions were generated by CAD of two nitrogen molecules from negatively charged phenyl tetrazole precursors. Their reactivities towards various reagents were explored and rationalized with the help of quantum chemical calculations.

Interface Balance Laws, Growth Conditions and Explicit Interface Modeling Using Algebraic Level Sets for Multiphase Solids with Inhomogeneous Surface Stress

Vaitheeswaran, Pavan Kumar Purdue University ProQuest Dissertations & Theses 2020 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Many natural and engineering phenomena including cancer cell growth, solidification, and crack propagation may be classified as phase evolution problems. These phenomena may have significant physiological or engineering impact, but are difficult to analyze due to the complexity of their interface geometry evolution and the complexity of their governing equations. The goal of this dissertation is to derive the thermodynamic conditions governing the evolution of the interfaces and an efficient computational procedure for modeling the interface evolution. The first contribution towards the goal is to derive the thermodynamic conditions at a moving interface in a body subject to large deformation with multiple diffusing species and arbitrary surface stress. A pillbox procedure is used to form balance laws at an interface, analogous to conventional balance laws in the bulk. The thermodynamic conditions that result from interface free-energy inequality lead to the analytical form of the configurational force for bodies subject to mechanical loads, heat and multiple diffusing species. The derived second law condition naturally extends the Eshelby energy-momentum tensor to include species diffusion terms. The above second law restriction is then used to derive the condition for the growth of new phases in a body undergoing finite deformation subject to inhomogeneous, anisotropic surface stress. Next, a general, finite-deformation, arbitrary surface stress form of phase nucleation condition is derived by considering uncertainty in growth of a small nucleus. The probability of nucleation is shown to naturally depend on a theoretical estimate of critical volumetric energy density, which is directly related to the surface stress. The classical nucleation theory is shown to result as a simplified special case.Towards the goal of computationally simulating evolving interfaces, an interface tracking approach called enriched isogeometric analysis (EIGA) is adopted in this thesis. The phase boundary is represented explicitly using parametric splines, with the physical fields isogeometrically defined on the interface geometry, and immersed in a non-conforming underlying domain. The behavioral field solution at a point is given by a convex blending of the underlying solution and the interface solution. Signed algebraic level sets are used as a measure of distance to model the weakening influence of a phase interface with distance. These level sets are generated from the implicitization of geometries using resultants, and are smooth, monotonic with distance, and exact on the boundary. Furthermore, the sign of these level sets enable classifying points as lying inside or outside a given closed geometry. The generation of these level sets is found to fail often for even simple three-dimensional surfaces, where the Dixon resultant used for implicitization is either identically zero or unsigned. A maximal-rank submatrix approach is adopted in this thesis to recover the implicitization for surfaces with identically zero resultants. Also, a polynomial square root procedure is developed to extract sign from unsigned resultants. The proposed approach is demonstrated on three-dimensional electrostatic and electromigration problems, and is used to simulate electromigration experiments conducted on Copper-TiN line structures. Since EIGA is an explicit interface method, topological changes that are common in phase evolution pose geometric challenges, such as computing intersection between boundaries of coalescent phases. This is overcome in this thesis by using Boolean compositions on algebraic level sets. These compositions can be done algebraically using R-functions, and provide level sets for merged phases.

Development of a Synthetic Method For 2-tetrazenes and Examination of the Mass Spectrometric Behavior of Ionized 2-Tetrazenes and Asphaltene Model

Lei, Haoran Purdue University ProQuest Dissertations & Theses 2022 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

2-Tetrazens are often used as high energy-density materials. Only a few synthetic methods exist for making them. Further, their structure-property relationships remain largely unexplored. Asphaltenes are usually a complex mixture found in the heaviest fraction of heavy crude oil. Their structural characterization and understanding have become a pertinent task for petroleum industry around the world. Mass spectrometry is a powerful analytical tool for the detection and characterization of unknown compounds even in complex mixtures. It features high sensitivity and speed and can provide a variety of valuable information for different types of analytes. This thesis focuses on the development of an effective synthetic method for 2-tetrazenes and the mass spectrometry study of the ionized 2-tetrazenes. In addition, the mass spectrometric behavior of ionized asphaltene model compounds are also discussed in this thesis.Chapter 2 describes the instrumentation and fundamental aspects of the mass spectrometers used in the research discussed in this thesis. Chapter3 introduces an effective synthetic method for 2-tetrazens based on iodine-mediated oxidative N-N coupling reaction of hydrazines. Chapter 4 presents the fragmentation behavior of ionized 2-tetrazenes in tandem mass spectrometry experiments. Upon collision-activated dissociation (CAD), ionized alkyl-substituted linear 2- tetrazenes underwent simpler fragmentation reactions than the ionized aromatic 2-tetrazenes or a cyclic 2-tetrazene. The observations were rationalized by using quantum chemical calculations. Chapter 5 presents the fragmentation behavior of ionized asphaltene model compounds under medium-energy collision-activated dissociation (MCAD). The comparison between the different behaviors of the molecular radical cations of these compounds provided useful information for the identification of related unknown compound.

The effects of including almonds in an energy-restricted diet on weight, body composition, visceral adipose tissue, blood pressure and cognitive function

Dhillon, Jaapna Purdue University ProQuest Dissertations & Theses 2016 해외박사(DDOD)

소속기관이 구독 중이 아닌 경우 오후 4시부터 익일 오전 9시까지 원문보기가 가능합니다.

Inclusion of almonds in an energy restricted diet has been reported to enhance or have no effect on weight loss. Their effects specifically on visceral fat stores during energy restriction have not been widely examined. Additionally, almond consumption has been associated with reduced blood pressure, but whether this is linked to or is independent of changes of body composition has not been examined. Moreover, almond consumption during energy restriction may be an effective strategy for reversing the negative effects of dieting on cognitive performance. The unique nutrient profile of almonds also has the potential to influence cognitive function post-prandially. The post-lunch dip in cognition is a well-established phenomenon of decreased alertness, memory and vigilance after lunch consumption and can be affected by lunch composition. Almonds which are higher in fat and lower in carbohydrate may be able to reduce this post lunch dip in cognition. Consequently, this dissertation had three primary aims. The first aim was to evaluate the effects of almond consumption as part of an energy-restricted diet on weight, visceral and subcutaneous adipose depots and blood pressure compared to a nut-free energy restricted diet. The second aim was to evaluate the effects of almond consumption as part of an energy-restricted diet on cognitive function. The third aim was to evaluate the acute effects of almond consumption on the post-lunch dip in cognitive function. A secondary objective of this dissertation was to develop an analytical approach to identify metabolic profiles associated with almond consumption to ascertain compliance in long term clinical trials. (Abstract shortened by ProQuest.).