In many applications, such as conversational agents, virtual reality, movies, and games, animated facial expressions of computer-generated (CG) characters are used to communicate, teach, or entertain. With an increased demand for CG characters, it is important to animate accurate, realistic facial expressions because human facial expressions communicate a wealth of information. However, realistically animating faces is challenging and time-consuming for two reasons. First, human observers are adept at detecting anomalies in realistic CG facial animations. Second, traditional animation techniques based on keyframing sometimes approximate the dynamics of facial expressions or require extensive artistic input while high-resolution performance capture techniques are cost prohibitive. In this thesis, we develop a framework to explore representations of two key facial expressions, blinks and smiles, and we show that data-driven models are needed to realistically animate these expressions. Our approach relies on utilizing high-resolution performance capture data to build models that can be used in traditional keyframing systems. First, we record large collections of high-resolution dynamic expressions through video and motion capture technology. Next, we build expression-specific models of the dynamic data properties of blinks and smiles. We explore variants of the model and assess whether viewers perceive the models as more natural than the simplified models present in the literature. In the first part of the thesis, we build a generative model of the characteristic dynamics of blinks: fast closing of the eyelids followed by a slow opening. Blinks have a characteristic profile with relatively little variation across instances or people. Our results demonstrate the need for an accurate model of eye blink dynamics rather than simple approximations, as viewers perceive the difference. In the second part of the thesis, we investigate how spatial and temporal linearities impact smile genuineness and build a model for genuine smiles. Our perceptual results indicate that a smile model needs to preserve temporal information. With this model, we synthesize perceptually genuine smiles that outperform traditional animation methods accompanied by plausible head motions. In the last part of the thesis, we investigate how blinks synchronize with the start and end of spontaneous smiles. Our analysis shows that eye blinks correlate with the end of the smile and occur before the lip corners stop moving downwards. We argue that the timing of blinks relative to smiles is useful in creating compelling facial expressions. Our work is directly applicable to current methods in animation. For example, we illustrate how our models can be used in the popular framework of blendshape animation to increase realism while keeping the system complexity low. Furthermore, our perceptual results can inform the design of realistic animation systems by highlighting common assumptions that over-simplify the dynamics of expressions.

To improve throughput, due date adherence, or tool usage in semiconductor manufacturing, it is crucial to model the duration of individual processes such as coating, diffusion, or etching. Equipped with such data, production planning can develop dispatch schemes and schedules for optimized material routing. However, just a few tools indicate how long a process will take. Many variables affect the runtime of tool recipes that are used to realize processes. These variables include wafer processing mode, historical context, batch size, and job handling. In this thesis, a model that allows inferring tool recipe runtimes with adequate accuracy shall be developed. Firstly, predictive models shall be built for selected tools with known runtime behavior to establish a baseline for the methodology. Tools will be selected to cover a broad spectrum of processing modalities. The main predictors will be revealed using variable importance analysis. Furthermore, the analysis shall reveal under which conditions recipe runtime modeling is most accurate. Secondly, a generic approach shall be created to model recipe runtime. By accounting for tool, process, and material context, methods would be investigated from feature selection and automatic model selection. Finally, a pipeline for data cleansing, feature engineering, model building, and metrics will be developed using historical data from a wide range of factory data sources. Finally, a scheme to operationalize the findings shall be outlined. In particular, this requires establishing model serving to enable consumption in applications such as dispatching or operator interfaces.

One of the crucial challenges for Data Center (DC) operation is inefficient thermal management which leads to excessive energy waste. The information technology (IT) equipment and cooling systems of a DC are major contributors to power consumption. Additionally, failure of a DC cooling system leads to higher operating temperatures, causing critical electronic devices, such as servers, to fail which leads to significant economic loss. Improvements can be made in two ways, through (1) better design of a DC architecture and (2) optimization of the system for better heat transfer from hot servers. Row-based cooling is a suitable DC configuration that reduces energy costs by improving airflow distribution. Here, the IT equipment is contained within an enclosure that includes a cooling unit which separates cold and back chambers to eliminate hot air recirculation and cold air bypass, both of which produce undesirable airflow distributions. Besides, due to scalability, ease of implementation, and operational cost, row-based systems have gained in popularity for DC computing applications. However, a general thermal model is required to predict spatiotemporal temperature changes inside the DC and properly apply appropriate strategies. As yet, only primitive tools have been developed that are time-consuming and provide unacceptable errors during extrapolative predictions. We address these deficiencies by developing a rapid, adaptive, and accurate hybrid model by combining a DDM and the thermofluid transport relations to predict temperatures in a DC. Our hybrid model has low interpolative prediction errors below 0.7 oC and extrapolative errors less than one half of black-box models. Additionally, by changing the studied DC configuration such as cooling unit fans and severs locations, there are a few zones with prediction error more than 2 oC. Existing methods for cooling unit fault detection and diagnosis (FDD) are designed to successfully overcome individually occurring faults but have difficulty handling simultaneous faults. We apply a gray-box model involves a case study to detect and diagnose cooling unit fan and pump failure in a row-based DC cooling system. Fast detection of anomalous behavior saves energy and reduces operational costs by initiating remedial actions. Cooling unit fans and pumps are relatively low-reliability components, where the failure of one or more components can cause the entire system to overheat. Therefore, appropriate energy-saving strategies depend largely on the accuracy and timeliness of temperature prediction models. We used our gray-box model to produce thermal maps of the DC airspace for single as well as simultaneous failure conditions, which are fed as inputs for two different data-driven classifiers, CNN and RNN, to rapidly predict multiple simultaneous failures. Our FDD strategy can detect and diagnose multiple faults with accuracy as high as 100% while requiring relatively few simultaneous fault training data samples. / Thesis / Candidate in Philosophy

Battery electric buses (BEBs) bring several advantages to public transportation systems. With fixed routes and scheduled trips, the implementation of BEBs in the transit context is considered a seamless transition towards a zero greenhouse gases transit system. However, energy consumption uncertainty is a significant deterrent for mainstream implementation of BEBs. Demonstration and trial projects are often conducted to better understand the uncertainty in energy consumption (EC). However, the BEB's energy consumption varies due to uncertainty in operational, topological, and environmental attributes. This thesis aims at developing simulation, data-driven, and low-resolution models using big data to quantify the EC of BEBs, with the overarching goal of developing a comprehensive planning framework for BEB implementation in bus transit networks. This aim is achieved through four interwind objectives. 1) Quantify the operational and topological characteristics of bus transit networks using complex network theory. This objective provides a fundamental base to understanding the behaviour of bus transit networks under disruptive events. 2) Investigate the impacts of the vehicular, operational, topological, and external parameters on the EC of BEBs. 3) Develop and evaluate the feasibility of big-data analytics and data-driven models to numerically estimate BEB's EC. 4) Create an open-source low-resolution data-based framework to estimate the EC of BEBs. This framework integrates the modelling efforts in objectives 1-3 and offers practical knowledge for transit providers. Overall, the thesis provides genuine contributions to BEB research and offers a practical framework for addressing the EC uncertainty associated with BEB operation in the transit context. Further, the results offer transit planners the means to set up the optimum transit operations profile that improves BEB energy utilization, and in turn, reduces transit-related greenhouse gases. / Thesis / Doctor of Engineering (DEng)

There has been significant growth in online social science experiments in order to understand behavior at-scale, with finer-grained data collection. Considerable work is required to perform data analytics for custom experiments. In this dissertation, we design and build composable and extensible automated software pipelines for evaluating social phenomena through iterative experiments and modeling. To reason about experiments and models, we design a formal data model. This combined approach of experiments and models has been done in some studies without automation, or purely conceptually. We are motivated by a particular social behavior, namely collective identity (CI). Group or CI is an individual's cognitive, moral, and emotional connection with a broader community, category, practice, or institution. Extensive experimental research shows that CI influences human decision-making. Because of this, there is interest in modeling situations that promote the creation of CI in order to learn more from the process and to predict human behavior in real life situations. One of our goals in this dissertation is to understand whether a cooperative anagram game can produce CI within a group. With all of the experimental work on anagram games, it is surprising that very little work has been done in modeling these games. Also, abduction is an inference approach that uses data and observations to identify plausibly (and preferably, best) explanations for phenomena. Abduction has broad application in robotics, genetics, automated systems, and image understanding, but have largely been devoid of human behavior. We use these pipelines to understand intra-group cooperation and its effect on fostering CI. We devise and execute an iterative abductive analysis process that is driven by the social sciences. In a group anagrams web-based networked game setting, we formalize an abductive loop, implement it computationally, and exercise it; we build and evaluate three agent-based models (ABMs) through a set of composable and extensible pipelines; we also analyze experimental data and develop mechanistic and data-driven models of human reasoning to predict detailed game player action. The agreement between model predictions and experimental data indicate that our models can explain behavior and provide novel experimental insights into CI. / Doctor of Philosophy / To understand individual and collective behavior, there has been significant interest in using online systems to carry out social science experiments. Considerable work is required for analyzing the data and to uncover interesting insights. In this dissertation, we design and build automated software pipelines for evaluating social phenomena through iterative experiments and modeling. To reason about experiments and models, we design a formal data model. This combined approach of experiments and models has been done in some studies without automation, or purely conceptually. We are motivated by a particular social behavior, namely collective identity (CI). Group or CI is an individual’s cognitive, moral, and emotional connection with a broader community, category, practice, or institution. Extensive experimental research shows that CI influences human decision-making, so there is interest in modeling situations that promote the creation of CI to learn more from the process and to predict human behavior in real life situations. One of our goals in this dissertation is to understand whether a cooperative anagram game can produce CI within a group. With all of the experimental work on anagrams games, it is surprising that very little work has been done in modeling these games. In addition, to identify best explanations for phenomena we use abduction. Abduction is an inference approach that uses data and observations. Abduction has broad application in robotics, genetics, automated systems, and image understanding, but have largely been devoid of human behavior. In a group anagrams web-based networked game setting we do the following. We use these pipelines to understand intra-group cooperation and its effect on fostering CI. We devise and execute an iterative abductive analysis process that is driven by the social sciences. We build and evaluate three agent-based models (ABMs). We analyze experimental data and develop models of human reasoning to predict detailed game player action. We claim our models can explain behavior and provide novel experimental insights into CI, because there is agreement between the model predictions and the experimental data.

Dissolved oxygen (DO) is an important water quality parameter that is used to assess the health of aquatic ecosystems. Typically physically-based numerical models are used to predict DO, however, these models do not capture the complexity and uncertainty seen in highly urbanised riverine environments. To overcome these limitations, an alternative approach is proposed in this dissertation, that uses a combination of data-driven methods and fuzzy numbers to improve DO prediction in urban riverine environments. A major issue of implementing fuzzy numbers is that there is no consistent, transparent and objective method to construct fuzzy numbers from observations. A new method to construct fuzzy numbers is proposed which uses the relationship between probability and possibility theory. Numerical experiments are used to demonstrate that the typical linear membership functions used are inappropriate for environmental data. A new algorithm to estimate the membership function is developed, where a bin-size optimisation algorithm is paired with a numerical technique using the fuzzy extension principle. The developed method requires no assumptions of the underlying distribution, the selection of an arbitrary bin-size, and has the flexibility to create different shapes of fuzzy numbers. The impact of input data resolution and error value on membership function are analysed. Two new fuzzy data-driven methods: fuzzy linear regression and fuzzy neural network, are proposed to predict DO using real-time data. These methods use fuzzy inputs, fuzzy outputs and fuzzy model coefficients to characterise the total uncertainty. Existing methods cannot accommodate fuzzy numbers for each of these variables. The new method for fuzzy regression was compared against two existing fuzzy regression methods, Bayesian linear regression, and error-in-variables regression. The new method was better able to predict DO due to its ability to incorporate different sources of uncertainty in each component. A number of model assessment metrics were proposed to quantify fuzzy model performance. Fuzzy linear regression methods outperformed probability-based methods. Similar results were seen when the method was used for peak flow rate prediction. An existing fuzzy neural network model was refined by the use of possibility theory based calibration of network parameters, and the use of fuzzy rather than crisp inputs. A method to find the optimum network architecture was proposed to select the number of hidden neurons and the amount of data used for training, validation and testing. The performance of the updated fuzzy neural network was compared to the crisp results. The method demonstrated an improved ability to predict low DO compared to non-fuzzy techniques. The fuzzy data-driven methods using non-linear membership functions correctly identified the occurrence of extreme events. These predictions were used to quantify the risk using a new possibility-probability transformation. All combination of inputs that lead to a risk of low DO were identified to create a risk tool for water resource managers. Results from this research provide new tools to predict environmental factors in a highly complex and uncertain environment using fuzzy numbers. / Graduate / 0543 / 0775 / 0388

Infrastructure deterioration has been attributed to insufficient maintenance budgets, lacking restoration strategies, deficient deterioration prediction techniques, and changing climatic conditions. Considering that the latter adds more challenges to the former, there has been a growing demand to develop and implement climate-informed infrastructure asset management strategies. However, quantifying the impact of the spatiotemporally varying climate metrics on infrastructure systems poses a serious challenge due to the associated complexities and relevant modelling uncertainties. As such, in lieu of complex physics-based simulations, the current study proposes a glass box data-driven framework for predicting infrastructure climate induced deterioration rates. The framework harnesses evolutionary computing, and specifically multigene genetic programming, to develop closed-form expressions that link infrastructure characteristics to relevant spatiotemporal climate indices and predict infrastructure deterioration rates. The framework consists of four steps: 1) data collection and preparation; 2) input integration; 3) feature selection; and 4) model development and result interpretation. To numerically demonstrate its utility, the proposed framework was applied to develop deterioration rate expressions of two different classes of concrete and steel bridges in Ontario, Canada. The developed predictive models reproduced the observed deterioration rate of both bridge classes with coefficient of determination (R2) values of 0.912 and 0.924 for the training subsets and 0.817 and 0.909 for the testing subsets of the concrete and steel bridges, respectively. Attributed to its generic nature, the framework can be applied to other infrastructure systems, with available historical deterioration data, to devise relevant effective asset management strategies and infrastructure restoration standards under future climate scenarios. / Thesis / Master of Applied Science (MASc)

[ES] Los modelos de riesgo en el ámbito de la salud son métodos estadísticos que brindan advertencias tempranas sobre el riesgo de una persona de sufrir un episodio adverso en el futuro. Por lo general, utilizan la información almacenada de forma rutinaria en los sistemas de información hospitalaria para ofrecer una probabilidad individual de desarrollar un resultado negativo futuro en un período determinado. Concretamente, en el campo de las enfermedades crónicas que comparten factores de riesgo comunes, los modelos de riesgo se basan en el análisis de esos factores de riesgo -tensión arterial elevada, glucemia elevada, lípidos sanguíneos anormales, sobrepeso y obesidad- y sus medidas biométricas asociadas. Estas medidas se recopilan durante la práctica clínica de manera periódica y, se incorporan a los modelos de riesgo para apoyar a los médicos en la toma de decisiones. Para crear modelos de riesgo que incluyan la variable temporal, se podrían utilizar técnicas basadas en datos (Data-Driven), de forma que se tuviera en cuenta el historial de los pacientes almacenado en los registros médicos electrónicos, extrayendo conocimiento de los datos en bruto. Sin embargo, en el ámbito de la salud, los resultados de la minería de datos suelen ser percibidos por los expertos en salud como cajas negras y, en consecuencia, no confían en sus decisiones. El paradigma Interactivo permite a los expertos comprender los resultados, para que los profesionales puedan corregir esos modelos de acuerdo con su conocimiento y experiencia, proporcionando modelos perceptivos y cognitivos. En este contexto, la minería de procesos es una técnica de minería de datos que permite la implementación del paradigma Interactivo, ofreciendo una comprensión clara del proceso de atención y proporcionando modelos comprensibles para el ser humano. Las condiciones crónicas generalmente se describen mediante imágenes estáticas de variables, como factores genéticos, fisiológicos, ambientales y de comportamiento. Sin embargo, la perspectiva dinámica, temporal y de comportamiento no se consideran comúnmente en los modelos de riesgo. Eso significa que el último estado de riesgo se convierte en el estado real del paciente. No obstante, la condición de los pacientes podría verse influenciada por sus condiciones dinámicas pasadas. El objetivo de esta tesis es proporcionar una visión novedosa del riesgo asociado a un paciente, basada en tecnologías Data-Driven que ofrezcan una visión dinámica de su evolución con respecto a su condición crónica. Técnicamente, supone abordar los modelos de riesgo incorporando la perspectiva dinámica y comportamental de los pacientes gracias a la información incluida en la Historia Clínica Electrónica. Los resultados obtenidos a lo largo de esta tesis muestran cómo las tecnologías de minería de procesos pueden aportar una visión dinámica e interactiva de los modelos de riesgo de enfermedades crónicas. Estos resultados pueden ayudar a los profesionales de la salud en la práctica diaria para una mejor comprensión del estado de salud de los pacientes y una mejor clasificación de su estado de riesgo. / [CA] Els models de risc en l'àmbit de la salut són mètodes estadístics que brinden advertències primerenques sobre el risc d'una persona de patir un episodi advers en el futur. Generalment, utilitzen la informació emmagatzemada de forma rutinària en els sistemes d'informació hospitalària per a oferir una probabilitat individual de desenrotllar un resultat negatiu futur en un període determinat. Concretament, en el camp de les malalties cròniques que compartixen factors de risc comú, els models de risc es basen en l'anàlisi d'eixos factors de risc -tensió arterial elevada, glucèmia elevada, lípids sanguinis anormals, sobrecàrrega i obesitat- i les seues mesures biomètriques associades. Estes mesures es recopilen durant la pràctica clínica ben sovint de manera periòdica i, en conseqüència, s'incorporen als models de risc i recolzen la presa de decisions dels metges. Per a crear estos models de risc que incloguen la variable temporal es podrien utilitzar tècniques basades en dades (Data-Driven) , de manera que es tinguera en compte l'historial dels pacients disponible en els registres mèdics electrònics, extraient coneixement de les dades en brut. No obstant això, en l'àmbit de la salut, els resultats de la mineria de dades solen ser percebuts pels experts en salut com a caixes negres i, en conseqüència, no confien en les decisions dels algoritmes. El paradigma Interactiu permet als experts comprendre els resultats, perquè els professionals puguen corregir eixos models d'acord amb el seu coneixement i experiència, proporcionant models perceptius i cognitius. En este context, la mineria de processos és una tècnica de mineria de dades que permet la implementació del paradigma Interactiu, oferint una comprensió clara del procés d'atenció i proporcionant models comprensibles per al ser humà. Les condicions cròniques generalment es descriuen per mitjà d'imatges estàtiques de variables, com a factors genètics, fisiològics, ambientals i de comportament. No obstant això, la perspectiva dinàmica, temporal i de comportament no es consideren comunament en els models de risc. Això significa que l'últim estat de risc es convertix en l'estat real del pacient. No obstant això, la condició dels pacients podria veure's influenciada per les seues condicions dinàmiques passades. L'objectiu d'esta tesi és proporcionar una visió nova del risc, associat a un pacient, basada en tecnologies Data-Driven que oferisquen una visió dinàmica de l'evo\-lució dels pacients respecte a la seua condició crònica. Tècnicament, suposa abordar els models de risc incorporant la perspectiva dinàmica i el comportament dels pacients als models de risc gràcies a la informació inclosa en la Història Clínica Electrònica. Els resultats obtinguts al llarg d'esta tesi mostren com les tecnologies de mineria de processos poden aportar una visió dinàmica i interactiva dels models de risc de malalties cròniques. Estos resultats poden ajudar els professionals de la salut en la pràctica diària per a una millor comprensió de l'estat de salut dels pacients i una millor classificació del seu estat de risc. / [EN] Risk models in the healthcare domain are statistical methods that provide early warnings about a person's risk for an adverse episode in the future. They usually use the information routinely stored in Hospital Information Systems to offer an individual probability for developing a future negative outcome in a given period. Concretely, in the field of chronic diseases that share common risk factors, risk models are based on the analysis of those risk factors -raised blood pressure, raised glucose levels, abnormal blood lipids, and overweight and obesity- and their associated biometric measures. These measures are collected during clinical practice frequently in a periodic manner, and accordingly, they are incorporated into the risk models to support clinicians' decision-making. Data-Driven techniques could be used to create these temporal-aware risk models, considering the patients' history included in Electronic Health Records, and extracting knowledge from raw data. However, in the healthcare domain, Data Mining results are usually perceived by the health experts as black-boxes, and in consequence, they do not trust in the algorithms' decisions. The Interactive paradigm allows experts to understand the results, in that sense, professionals can correct those models according to their knowledge and experience, providing perceptual and cognitive models. In this context, Process Mining is a Data Mining technique that enables the implementation of the Interactive paradigm, offering a clear care process understanding and providing human-understandable models. Chronic conditions are usually described by static pictures of variables, such as genetic, physiological, environmental, and behavioural factors. Nevertheless, the dynamic, temporal, and behavioural perspectives are not commonly considered in the risk models. That means the last status of the risk becomes the actual status of the patient. However, the patients' condition could be influenced by their past dynamic circumstances. The objective of this thesis is to provide a novel risk vision based on Data-Driven technologies offering a dynamic view of the patients' evolution regarding their chro\-nic condition. Technically, it supposes to approach risk models incorporating the dynamic and behavioural perspective of patients to the risk models thanks to the information included in the Electronic Health Records. The results obtained throughout this thesis show how Process Mining technologies can bring a dynamic and interactive view of chronic disease risk models. These results can support health professionals in daily practice for a better understanding of the patients' health condition and a better classification of their risk status. / Valero Ramón, Z. (2022). Dynamic Risk Models for Characterising Chronic Diseases' Behaviour Using Process Mining Techniques [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/181652