Resource-Aware Distributed Analytics and Machine Learning for Hybrid Edge-Cloud Systems

Das, Anirban ProQuest Dissertations & Theses Rensselaer Polytec 2021 해외박사(DDOD)

With more intelligent applications, data analytics and inference at the edge are proliferating as a complement to traditional computation done at a centralized cloud location. At the same time, distributed machine learning training at the edge of the network near the data producers is also gaining popularity, mainly due to benefits to security, privacy, and communication costs. However, edge devices are often resource-constrained, and further, there may be communication bottlenecks between the edge and the cloud. Successful solutions for these edge computing workloads must address challenges posed by constrained computation and communication resources.The first part of the thesis focuses on scheduling and task placement of data processing, analytics, and inference workloads. The goal is to provide some quality of service, for example, low latency or cost reduction in the context of edge-cloud architectures. We start with benchmarking the leading industry edge computing platforms that use the serverless computing paradigm as the medium of execution. We next consider serverless applications, consisting of a single-stage, and propose a framework to jointly execute such applications in the presence of an edge device and the public cloud. The aim is to decide whether to execute user jobs at the edge or the public cloud based on given latency or cost constraints. Finally, we consider a hybrid cloud scenario, where we consider a private cloud instead of a single edge device. Here, we study the problem of task placement and scheduling of multi-stage serverless applications between a private and the public cloud to minimize the cost of public cloud usage. In the second part of the thesis, we consider machine learning training workloads in edge-cloud platforms. More specifically, we study federated learning in this part of the thesis. Like the first part of the thesis, we first conduct a feasibility study of federated learning algorithms on resource-constrained devices. Next, we study an algorithm for horizontal federated learning in a hierarchical communication network. We analyze the convergence of the algorithm when there is a non-IID data distribution among the participants. Our analysis shows that the non-IID data distribution can have a significant impact on the algorithm convergence error. This insight paves the way for a more sophisticated algorithm design to diminish this performance gap. We then turn our focus towards vertical federated learning in a hierarchical network. We propose a new algorithm for model training where data is vertically partitioned across silos in the top tier and horizontally partitioned in the bottom tier among clients inside each silo. We present a theoretical analysis of our algorithm and show the dependence of the convergence rate on the number of vertical partitions, the number of local updates, and the number of clients in each hub.Lastly, we close with the summary and discussions on the future research directions and open questions of interest.

Quantitative Temperature Sensing and Thin Film Thermal Conductivity Measurement by Non-Contact Scanning Thermal Microscopy

Zhang, Yun ProQuest Dissertations & Theses Rensselaer Polytec 2020 해외박사(DDOD)

Scanning Thermal Microscopy (SThM) is a powerful technique that can measure samples’ thermal conductivity, temperature, Seebeck coefficient, and topography at the same time with excellent spatial resolution and sensitivity. It has a lot of potential applications in material science, nanoelectronics design, and even biological field. The current state-of-the-art SThM focuses on the contact mode that has inaccuracy due to surface artifacts and suffers from probe damage. Therefore, this thesis reports the investigation on the development, validation, and application of a quantitative non-contact SThM under ambient conditions for temperature sensing and high thermal conductivity 2D sample characterization. This thesis is divided into six chapters. The first chapter introduced the history and background of the SThM development and states the motivation of this thesis. The second chapter describes the experimental, numerical, and analytical methods for a proof-of-concept demonstration. The experiments involve a passive probe that measures sample temperature and an active that measures sample thermal conductivity in the diffusive and the transition regimes. The analytical models of the active and passive probes and a multilayer sample are also introduced. A cutting procedure forces the 3-Dimensional Finite Element Model (3DFEM) to take the transition heat transfer between the probe and the sample into account and a useful correlation of the probe-sample air gap thermal resistance is developed by the validated 3DFEM. Then the discussion continues to the calibration of the thermal exchange parameters, which are important for the quantitative and accurate temperature and thermal conductivity results. The probe-sample air gap thermal resistance correlation together with the analytical models forms the basis of the active and passive calibration techniques for temperature sensing. For film thermal conductivity measurement, the thermal exchange radii are found by fitting the 3DFEM heat flux profile on the sample surface. Chapter 3 reports the validated results and demonstrates the experimental setup having a 700 nm spatial resolution and 0.01 K temperature resolution of the temperature sensing and a film thermal conductivity of up to ~240 W/(m·K) and down to 0.2 W/(m·K) with thickness from 46.6 nm to 240 nm can be measured with the setup. Extended on the validated 3DFEM, Chapter 4 discusses the definitions and methods to evaluate the sensitivity and spatial resolution of commercial thermoresistive probes. The sensitivity and spatial resolution of the Wollaston wire probe were also documented in detail when measuring film thermal conductivity. Generally, a probe with a more focused heat source will have a higher spatial resolution and a probe with a more spreading heat source will have a better sensitivity. The existence of the cantilever may significantly affect the overall performance. The DS probe was found to have the best sensitivity and the Nanowire probe had the finest spatial resolution when they are measuring thermal conductivities. The Wollaston wire probe had the highest spatial resolution and the DS probe is most sensitive to the sample temperature change. The investigation of the probe performances inspires the study of modified probe geometry for the KNT and the Wollaston wire probe and the optimized Wollaston wire probe has a 2.4 μm diameter. Chapter 5 summarizes this thesis and highlights important conclusions. The unaddressed challenges of parasite effect of temperature sensing and further probe optimization and fabrication are emphasized as guidance for future research.

Probabilistic Acousto-Ultrasonic Active-Sensing Structural Health Monitoring Based on Gaussian Process and Stochastic Time Series Models

Amer, Ahmad ProQuest Dissertations & Theses Rensselaer Polytec 2021 해외박사(DDOD)

In the context of engineering structures, structural safety, maintenance and life-cycle management processes are a major factor in sustainability. In particular, the aerospace industry is one that depends heavily on schedule-based procedures in order to sustain proper life-cycle management, ensure safety, and improve performance. Most of such procedures include some type of Non-destructive Evaluation (NDE) techniques, in which aircraft need to be inspected on a regular basis on the ground before operations can be resumed regardless of structural state. This framework, although very effective in the sustainability efforts of the aerospace industry, suffers from a number of drawbacks; the most economically-prominent of which are cost, increased downtime, less-than-optimal safety management paradigm (damage can occur and grow between scheduled procedures) and the limited applicability of fully-autonomous operations. As such, research endeavors in the past 40 years have been directed towards developing sustainability efforts that can be applied online (limiting downtime and increasing safety) and in an automated fashion (limiting the need for costly man hours, and also allowing for autonomous operation). Aside from the implementation of such frameworks in the industry of rotating machinery, the collection of these online frameworks falls under the field of Structural Health Monitoring (SHM).Because of the complexity of aircraft operations, manifested in multiple operational cycles, and, within each cycle, the varying operational and environmental conditions, the aerospace industry poses as a very rich arena for development of SHM techniques \\cite{Dong-Kim18}. When it comes to active-sensing guided-wave SHM in particular, where piezoelectric sensors communicate with each other, owing to the fact that most of the currently-employed approaches are of a deterministic nature, i.e. they do not account for operational, environmental and modelling uncertainties, the complexity of aerospace SHM creates a number of challenges in the face of researchers in the active-sensing SHM field today. Namely, emerging SHM technologies need to be accurate and robust in the face of stochastic time-varying and non-linear structural responses, as well as incipient damage types and complex failure modes that can be easily masked by the effects of varying operational and environmental conditions. In addition, with the advancement in on-board data acquisition technologies, SHM frameworks need to be data-intelligent i.e. they need to be capable of using data efficiently.Thus, there lies a need for the development of active-sensing SHM frameworks, where proper understanding, modeling, and analysis of stochastic structural responses under varying states and damage characteristics is achieved for clearing the road towards achieving the aforementioned ultimate goal of SHM systems. This is where probabilistic SHM comes in. This thesis attempts to pave the way towards fully-probabilistic frameworks for active-sensing, guided-wave SHM. With focus on damage detection and quantification, statistical and probabilistic techniques are put forward that not only properly model uncertainties in the data coming from the system being interrogated, but also surpass currently-used methods in accuracy. In addition, this thesis addresses the issue of data-intelligence of probabilistic models through a number of approaches.The first problem tackled in this thesis is statistical damage detection, where statistics based on non-parametric time series representations are proposed and applied to test cases to compare their detection performance with standard state-of-the-art damage indicators. Then, once damage is detected, the problem of data-intelligence is addressed through proposing a statistical signal path selection algorithm, again based on non-parametric time series models, which classifies signal paths into damage-intersecting and non-intersecting, where only the former is used in training damage quantification models. After that, probabilistic damage quantification is addressed next through proposing three frameworks based on the probabilistic machine learning techniques within the family of Gaussian Process (GP) models. The first probabilistic damage quantification framework uses industry-standard damage indicators to build the GP models. The second GP framework uses one of the damage detection statistics mentioned above. The third quantification framework uses GP models that are trained using time-varying parametric time series representations. Other work don in this thesis include the integration of physics-based load-compensation models with GP models for situations where critical data is missing in the training space. Also multi-output GP models are proposed to leverage information from across a sensor network for better damage quantification.All in all, the methods presented in this thesis are intended to bring the active-sensing, guided-wave SHM community one step closer to a full probabilistic treatment of SHM problems; the research herein should be considered as a stepping stone towards that goal. This being said, from the studies conducted in this thesis, a plethora of open research questions that need to be answered emerge, with some of them mentioned at the end of the thesis.

Towards Automated Axiom Generation: A Semi-Automated Approach to Generating "Knowledge and Rule Base" Corpora from Text Narratives

Prabhu, Anirudh ProQuest Dissertations & Theses Rensselaer Polytec 2021 해외박사(DDOD)

With the exponential rise of data in recent years, deep learning has risen to be one of the most prominent forms of artificial intelligence. With many successful applications, deep learning has helped researchers build machines that successfully complete human tasks previously thought to be very difficult. For example, restoring color to black and white photos, image captioning, voice generation, restoring sound in silent videos, lip reading from videos etc. are some very interesting applications being explored and with deep learning. Even with success in a breadth of applications, there are still problems that deep learning has not been able to solve. For example, scalability, understanding context, or examining and understanding the inner workings of deep neural networks themselves remain unsolved problems. The crux of the deep learning approach are “layers” (input, hidden and output). These layers are adjustable to a given corpus and mostly opaque to interpretation or explanation. Current approaches to Artificial Intelligence/Machine Learning rely on an entire corpus, i.e. they use the entire content with noise, bias, etc. These approaches have achieved high success across fields like computer vision, natural language processing, image captioning etc., require a very large amount of training data to accurately understanding the mappings between the input and output embeddings for the deep learning experiment. In this thesis, we ask the question, "What if, intelligent information extraction (both entity and relation) were able to provide a curated corpus for deep network learning?" Curation in this context, addresses eliminating all the "non-essential" parts of text, and simply focusing on the actions, agents and events involved in a text corpus, and the rules that highlight the effects of these actions and change in the narrative.What is needed to achieve this task, is the ability to recognize key entities and map the situational changes occurring in the corpus to specific triggers (such as actions or events), like those seen in axioms in a rule base. Automated axiom creation is a difficult research problem to solve. Most of the work in this area focuses on rules extracting rules from text that explicitly mentions the rule in text. In most old fashioned AI systems, rules are developed with an understanding of the domain and reading between the lines where required to see what action/events could trigger a particular response in the narrative. An automated axiom creation method that completes such a task is still an unexplored and unsolved problem, and the focus of this thesis.The biggest hurdle in exploration of this research problem, is the availability of data (or the lack thereof) where implicit rules are documented for a text narrative. In this thesis, we have developed a novel semi-automated method to generate axioms/rules for a set of text narratives, using crowdsourcing and known natural language processing techniques. We begin with textual narrative such as those in novels, computer manuals, but also in view are scientific works. We then document rules for the given narrative by using Amazon Mechanical Turk, a crowdsourcing platform known to aid in the creation of high-quality datasets. We have found that the usage of a crowdsourcing platform works well for narratives that do not require any expertise, like those seen in novels, and are able to provide textual rules for the narrative which may not be explicitly stated in the text. The next step would be to process these narratives and their rules into knowledge bases and rule bases, where the key concepts and relationships need to be extracted from both the knowledge bases (in the form of triples) and rule base. We have also developed an approach to converting the results of the crowdsourcing experiment (rules in text form), to formal rules in a rule base. These are developed based on known NLP information extraction techniques, like POS tagging, co-reference resolution etc. The overall goal of this thesis is a novel method to extract key information and rules from a narrative , in order to create a set of knowledge bases and rule bases. After examining the results of the crowdsourcing experiment, we found a set of boundaries for the usability of crowdsourcing as tool or means to overcome the automation bottleneck. We also discuss the required distinction of the terms "Humans in the loop" and "Experts in the loop", and provide a platform for fleshing out the framework for experts in the loop for scientific workflows. Finally we also developed a method to evaluate "knowledge base—rule base" corpora for any logical language in any domain.To construct such a "situational narrative", a formalism such as situation calculus stands out as an obvious choice for knowledge representation but is heretofore an unexplored option in explaining “what is going on” in deep learning. At the heart of such a capability may be a learned formalization of the “situation” and perhaps even the identification of changes in state or fluent(s) (situation) over iterations or after learning interactions.

Data-Driven Methods for Probabilistic Fault Diagnosis in Multirotor Aircraft

Dutta, Airin ProQuest Dissertations & Theses Rensselaer Polytec 2022 해외박사(DDOD)

The next revolution in aviation is upon us with advances in novel configurations of electric vertical take-off and landing (eVTOL) aircraft. Advanced air mobility (AAM) will offer on-demand services for human and cargo transport and package delivery in the major cities of the world. Its other anticipated uses include surveillance for public safety, humanitarian aid, infrastructure supervision, remote sensing, etc. However, the operational success of mass transportation services by aerial vehicles will require absolute safety and reliability making efficient health and usage monitoring (HUMS) of these systems vital. According to a technical report by Uber Elevate, the safety level in air-taxi aviation needs to improve from 1.2 to 0.3 fatalities per 100 million passenger miles through full autonomy and innovation, with large amounts of data from real-world operations after the first generation VTOL aircraft are in production. Therefore, research and development are imperative to realize real-time system-level awareness and decision-making, in future intelligent and autonomous VTOL aircraft.This line of work focuses on fault detection and identification (FDI) of system faults in potential AAM vehicles utilizing in-flight data streams. Knowledge of system faults in real-time is critical for control reallocation or vehicle reconfiguration to complete the flight safely. Moreover, the incorporation of condition monitoring from the early phases of eVTOL operation will boost aircraft readiness, enhance flight safety, and lower maintenance, and operating costs. These will ensure the commercial success of the large fleet of frequently flying aerial vehicles. There has been extensive research going on to implement fault-tolerant control on multirotor aircraft, most of which relies on information about system faults in real-time to switch onto more power-efficient optimum control schemes or plan alternate trajectories with limited control authority awareness, depending upon the type and extent of faults. However, the analytical FDI approaches are mostly limited by the requirement of in-depth physical knowledge of the aircraft and lack of efficient handling of noise, and uncertainty, while the data-driven approaches suffer from a lack of explainability due to focusing on fitting the data and concentrate mostly on structural faults in blades and powertrain components of single rotor platforms. Therefore, the research gap related to probabilistic actuator FDI has been explored in this thesis. In this study, the following challenges pertaining to the development of a probabilistic multicopter FDI technique have been addressed. First, it should be robust under operational variability, environmental disturbances, and uncertainty. Second, online fault monitoring should be made possible by improving the run-time of the decision-making scheme through low-dimensional representations of the dynamic information contained in the multi-modal sensory data. Third, these low-dimensional representations must be physically explainable, based on stochastic representations of the multicopter dynamics contained in data streams (aircraft states and controls time-series data). Therefore, in the context of probabilistic FDI, a stochastic framework for FDI in multicopters is proposed, which attains the goal of being accurate, robust, and data-driven with improved physical interpretability. Its cornerstone lies in “global” stochastic time-series models which can appropriately represent the dynamics of the aircraft flight signals under multiple flight states, different fault types and magnitude, changing environmental disturbances, and uncertainty via functional pooling of data. At first, residual-based statistical time-series methods have been investigated with a novel application to multicopter rotor FDI. Some of these methods exhibit excellent accuracy but suffer from certain limitations that have been addressed through an innovative approach that integrates statistical time-series modeling and a machine learning algorithm. This method, titled the time-series assisted neural network has the following advantages over the former: (i) it requires only the healthy stochastic model to derive fault-sensitive (type and magnitude), and disturbance-rejecting features, (ii) it makes probabilistic decisions regarding the rotor faults in a single step using a simple neural network, (iii) it is applicable throughout the entire flight regime and has better accuracy with shorter signals enabling faster FDI.In the second part of this thesis, flexible booms have been incorporated into the multicopter to generate simulated data. This opened new avenues for signal selection from remote and local sensors to achieve better uncertainty quantification in rotor fault magnitude estimation. It was achieved via inverse optimization techniques with the aforementioned “global” stochastic models representing the functional dependence of signal dynamics with varying fault magnitude. Exploring local sensors mounted on the booms also led to the development of a probabilistic rotor fault diagnosis framework based on simple machine learning algorithms. It was developed using out-of-plane strain signals at individual boom roots and exhibited over 99% rotor FDI accuracy under any admissible operating conditions and external disturbances without the need for dynamic representations or knowledge of the operating states. In the final task, the time-series assisted neural network performance was validated with experimental data from flight tests of a quadcopter and a hexacopter.

Investigations of the Aptamer Capability of G-Quadruplex-Forming Oligonucleotides

Albanese, Christina M ProQuest Dissertations & Theses Rensselaer Polytec 2016 해외박사(DDOD)

The unrelenting demand for affinity reagents in areas such as proteomics, medical diagnostics, and therapeutic treatment has necessitated their further study and development. In recent years, aptamers---short, single-stranded oligonucleotides that bind target molecules with high affinity and specificity---have become increasingly attractive candidates. Since the introduction of the Systematic Evolution of Ligands by Exponential Enrichment (SELEX) process in 1990, numerous aptamers have been identified. Despite extensive efforts towards identification of aptamers to new targets, progress has been hindered by the limitations of the SELEX method. Therefore, finding other avenues for aptamer discovery is crucial for further advancement of this field. This dissertation first investigates a novel, genome-inspired reverse-selection pathway towards aptamer discovery that was previously introduced in our laboratory. In this approach, genome-inspired DNA sequences are selected as potential aptamers that may specifically bind proteins extracted from human cells. Selectively captured proteins are identified and the DNA-protein binding interactions are analyzed in live cells to determine if the interaction may have biological as well as analytical significance. Previous success in our group with the ERBB2 breast cancer promoter region encouraged expansion of our studies to include other G-rich promoter regions, which is the topic of this dissertation. These studies focus on the G-quadruplex-forming sequences from the c-myc, Rb, and VEGF oncogene promoter regions and their interactions with nuclear and cellular proteins from breast cancer cell lines. We were able to identify several proteins that bind in vitro to the G-quadruplex structures formed by these sequences, which will lead to new aptamers to these protein targets. Studies of these binding interactions in live cells through chromatin immunoprecipitation (ChIP) also indicate that these interactions occur in the chromatin of live cells. Since these oncogenes are overexpressed in many cancers, such binding could play a role in unearthing new cancer biomarkers or drug therapies. The second focus of this dissertation explores the selectivity and diagnostic potential of the G-quadruplex-forming VEGF aptamer. This aptamer selectively binds the VEGF protein, which is present at high serum levels in numerous cancers and diseases. Using an affinity MALDI MS platform, we covalently attached the VEGF aptamer to fused silica MALDI probe surfaces to selectively capture the VEGF protein from a complex mixture of human serum proteins. These results demonstrate the value of our affinity MALDI method and may allow for rapid protein screening of diseases characterized by high VEGF serum levels. Together, these studies demonstrate the potential of G-quadruplex-forming oligonucleotides, both for use as aptamers and as possible gene regulatory elements in cancer and other diseases.

Post-Damage Reconfiguration for Rotorcraft with Control Redundancy

Vayalali, Praneet ProQuest Dissertations & Theses Rensselaer Polytec 2021 해외박사(DDOD)

Aircraft survivability in the event of component failure or some loss of control effectiveness is a critical area of research, particularly with regards to the control system design. While this has been thoroughly researched for fixed-wing aircraft since the 1980s, there has been a lack of a similar body of work for rotorcraft. This lack is mainly due to the absence of control redundancy on a conventional single main rotor helicopter. A fully compounded helicopter bridges this gap by adding fixed-wing control surfaces (flaps, ailerons, stabilator, and rudder) and compound-specific auxiliary controls (propeller thrust and main rotor speed). When the platform of interest possesses a significant amount of control redundancy, the allocation of these controls plays a vital role in the system's performance and longevity. The design choices made can allow for tolerance to a range of different control failures. The availability of redundant control effectors allows for an exploration of its ability to tolerate damage on an aircraft, thereby improving its survivability. Furthermore, there is little understanding of a rotorcraft's flight dynamics and transient behavior when damage occurs, and the controls are reconfigured to tolerate such damage. This exploration forms the crux of this dissertation.On a UH-60 Black Hawk, the stabilator can act as a redundant control effector at moderate to high-speed flight conditions. A steady-state trim analysis is performed to demonstrate the feasibility of trimmed flight in various conditions with different locked servo actuator positions for the forward, aft, and lateral actuators. After failure, the controls are reconfigured to partially reallocate the control authority in the longitudinal axis from the main rotor longitudinal cyclic to the stabilator. Flight simulation results demonstrate the ability of this reallocation to compensate for locked-in-place failure of the forward main rotor swashplate servo actuator, as well as the ability of the aircraft to recover safely through a rolling landing maneuver. A similar range of locked positions of main rotor swashplate actuators is demonstrated to be feasible for aircraft recovery using control of the stabilator. So far, stabilator use has been shown to work in an adaptive sense, where the control mixing is remapped in flight once failure is detected. Next, it is shown to perform well when the defined mixing utilizes the stabilator even on the undamaged aircraft, removing the need to detect and identify specific failures on the aircraft. Further investigation considered the benefit of allowing for more or less longitudinal authority to be given to the stabilator in different flight conditions in the context of handling qualities ratings for the aircraft in pitch attitude and vertical rate response. Stabilator hardover failure is also examined. This work is then extended onto the compound helicopter platform. A reconfigurable control allocation method is applied on a compound helicopter in order to utilize the redundant control effectors in the feedback loop to compensate for locked-in-place actuator failures. A range of tolerable positions for locked-in-place actuator failures is established for the aircraft at a cruise speed of 150 knots. A full authority model following linear dynamic inversion control architecture is implemented for the nonlinear simulation model. It is shown to successfully compensate for actuator failures when the feedback control and the pseudoinverse control allocation method redistributes the control authority to the working actuators, assuming fault detection has taken place. Finally, a comparison of the robustness of a baseline pseudoinverse control allocation to an adaptive redistributed pseudoinverse method when the aircraft is subjected to different actuator failure at their extreme positions is examined. This is carried out through handling qualities based assessment and dynamic nonlinear flight simulations.

Linking Adaptation Processes to Team Performance in High-Tempo, High-Stakes Teamwork: A Large-Scale Gaming Perspective

Eaton, Joshua A. N ProQuest Dissertations & Theses Rensselaer Polytec 2020 해외박사(DDOD)

Through four interrelated studies, this dissertation examines behavioral adaptations within teams in response to planned-for and unplanned-for contingencies in extreme task environments, whether precipitated by changes within the team itself, or through a team’s operational experience. This dissertation is unique in its focus on highly detailed, naturally occurring data collected longitudinally from thousands of teams. The data are taken from the Multi-player Online Battle Arena game, League of Legends (LoL). The Input-MediatorOutcome Team Effectiveness Model provides the organizing framework for this dissertation.The first study investigates the effect of losing a “critical” team member (CTM) on the performance of teams. Statistical analysis is used to explore the effects on team performance. Initial results indicate that the presence or absence of a CTM has a significant impact on team processes and performance outcomes. This research illustrates the feasibility of exploring online gaming data for new insights into team performance, leading to a larger study on team effectiveness. The second study evaluates the effect of two inputs (team familiarity and CTM) on team effectiveness. Statistical data modeling techniques are used to explore these relationships. Results from this study are used to motivate further work on exploring possible mediators involved in this relationship, as discussed in the descriptions of the following two studies.The third study explores the impact of one input (role familiarity) on a multi-dimensional measure of team performance as output (performance effectiveness and efficiency). This study hypothesizes that the importance of role familiarity is mediated by the nature and extent of team members’ experience working together in defined roles. ANOVA and visualization techniques are used to explore match-level data in order to address the proposed research questions. Finally, the fourth study examines the effect of team familiarity, role familiarity, and CTM on team performance as mediated by adaptive behaviors in executing teamwork processes through a fully tested integrated team effectiveness framework. The interrelated studies demonstrate that generalizations of this modeling approach can be applied to other domains where the quantification of individual and team-level workflow, role familiarity, and team familiarity are important to understanding performance.

Modeling the Viscosity and Solubility of Protein Solutions

Virk, Sabitoj Singh ProQuest Dissertations & Theses Rensselaer Polytec 2022 해외박사(DDOD)

Viscosity and solubility are one of the most important physical properties integral to the manufacturing and use of nearly every consumer product. The formulation of materials ranging from toothpastes to therapeutic drugs must be carefully optimized to maintain the structural intactness and flow characteristics necessary for the application of interest. Therefore, accurately modeling both of these properties as a function of solution conditions can serve as a powerful method to engineer a wide range of materials. A major issue faced during the manufacturing and administration of therapeutic protein formulations is the undesirable high viscosity. Numerous models have been developed to predict the viscosity of protein solutions but many fall short of the task due to an incomplete fundamental understanding of the underlying protein-protein interactions. Can an accurate depiction of the protein-protein interactions be the missing key to successfully model the viscosity of protein solutions? It is the need for an answer to this question that led to the research presented in this dissertation. The effect of the microscopic particle-particle interactions on the macroscopic solution viscosity is understood in this work by building a simple hard sphere colloidal model. Employing a particular combination of short-ranged attractions and long-ranged repulsions (SALR interactions), the zero-shear viscosity and the osmotic second virial coefficient for a dilute colloidal suspension were computed as a function of the attractions and the repulsions strength. A major finding showed that the long-ranged repulsions (LR) had a dual action on the suspension zero-shear viscosity. The long-ranged repulsion interactions (LR) acting alone always increased the viscosity but when combined with the short-ranged attractions (SALR) reduced or lowered the viscosity of the system. The analytical approximations were written to deduce the coupling mechanism of the interactions and their effect on reducing the suspension viscosity. This new knowledge shall be very helpful for designing protein formulations with low viscosity.With this novel insight, the simple hard sphere colloidal model with SALR interactions was applied towards predicting the viscosity of dilute to semi-dilute protein solutions. The comparison was performed for a globular shaped albumin and Y shaped therapeutic monoclonal antibody that were not explained by the previous colloidal models. The model predictions showed that it was the coupling between attractions and repulsions that gave rise to the observed experimental trends in solution viscosity as a function of pH, concentration, and ionic strength. The model was subsequently applied to assess the viscosity risk of a large panel of antibody candidates chosen for therapeutic development. Using the measurements of two high throughput in-vitro early-stage screening assays, the model acted as an attractive tool for predicting the developability assessment for antibodies.The thesis also touches upon the influence of SALR interactions on the protein solubility. It does so by modeling the synergistic precipitation of bovine serum albumin from its solution by using a combination of two precipitants. The mechanism of synergy in the solubility-reducing action of precipitants is not clear and has been a standing problem in the field of protein purification by precipitation. The work shows that the synergy occurs due to the coupling between the microscopic protein interactions.The belief is that this dissertation shall propel fellow scientists and engineers to think about the macromolecular solution properties in terms of the particle interactions happening at the microscopic scale.

Towards the Computation Problems in Multi-Stage and Multi-Winner Voting Rules

Wang, Jun ProQuest Dissertations & Theses Rensselaer Polytec 2022 해외박사(DDOD)

Multi-stage and multi-winner voting rules are playing an increasingly important role in society. The former consists of a large number of various procedures of multiple rounds based on repeated ballots and/or sequential elimination, such as Single Transferable Vote (STV), Baldwin rule, Coombs rule, Ranked Pairs (RP), etc., while the latter can select multiple candidates to win election to an office. For a long time, they have been well studied for their properties in different axioms, their manipulability, and the computational complexity of calculating a manipulation.As for the multi-stage voting rules, however, the literature is surprisingly vague about the impact of tiebreaking. Unlike previous default handling methods like alphabetical tiebreaking, our goal focuses on computing the set of all possible winners under any tiebreaking mechanism, henceforth known as parallel-universes tiebreaking (PUT). In this research, multiple algorithms are proposed based on depth-first search together with pruning strategies, heuristics, sampling, and machine learning which prioritizes search direction and substantially improves the performance. Also, novel integer linear programming (ILP) formulations are proposed for PUT-winners under STV and Ranked Pairs as a comparison, and the experiments' results show that the search algorithms are overall faster than ILP.Besides, much of this dissertation focuses on the margin of victory (MOV) problem. Margin of victory, concisely defined as the minimum number of ballots needed to be changed to alter the outcome, is another widely studied issue in voting theory. It is a crucial measurement for robustness of election outcome, which is very difficult and time-consuming to compute for multi-stage voting rules. To address this, this dissertation proposes an efficient algorithm for the exact MOV of STV computation, which has a significant time reduction compared to the current best-known Blom et al.'s approach. And this research first provides an efficient search algorithm for exact MOV of ranked pairs with the help of linear programming as a basic tool and machine learning as an accelerator technique.In addition, this dissertation also investigates multi-winner voting rules experimentally and exploratively. Contrary to the approximation algorithms discussed in most of the literature so far, this dissertation proposes the first search algorithms for solving the exact committee of winners for Chamberlain-Courant rule, where the machine learning aided prioritization proves to be valid again in early discovery. Also, this dissertation proposes an integer linear programming formulation which is the first for computing the exact MOV under the Chamberlain-Courant rule.