Title Promoter Affiliations Abstract "Developing statistical models for multivariate left-censored variables: a joint model using copula functions for the association and a conditional regression model with left-censored response and covariate variables." "Roel BRAEKERS" "Centre for Statistics" "Left-censored data is commonly obtained when a measuring device is not able to measure the variables of interest below certain tresholds. In this project, we introduce first a joint model for a bivariate vector of left-censored data by using a copula function for the association between the components. For the marginal distribution of each variable we assume a semi-prametric Cox's regression model. Afterwards, we develop a conditional regression model in which both the response and the covariate variable are left-censored. In this model, the distribution of the covariate variable has a strong influence on the finite dimensional parameters of the conditional regression model. To minimize this influence, we assume first a nonparametric Kaplan-Meier estimator for the left-censored covariate and afterwards a smoothed version of it through Bernstein polynomials. As results of this project, we investigate the asymptotic consistency and asymptotic normality of the finite dimensional parameters in the different models. Hereto we also show the almost sure consistency of the infinite dimensional parameters in the form of the baseline hazard function of the Cox's regression model or the non-parametric Kaplan-Meier estimator for the left-censored covariate. Furthermore these methods are illustrated through on real life data sets and userfriendly oftware will be provided." "beurs EVDS Arne Peirsman: Design of a 3D printed and perfused breast cancer metastasis model (3D PAP Be CAME model)" "Olivier De Wever" "Department of Human Structure and Repair" "The research project focusses on designing a model that mimics breast cancer metastasis. Today, 90% of breast cancer (BC) death is cased be metastasis. This is largely because we are still unable to understand the mechanisms of metastatic progression. Consequently, we are unable to drug and treat to cure metastatic disease. Better models of cancer metastasis will lead to more fundamental insights. Drug evaluation using this model will enable clinicians to make more rational decisions for drug efficiency. In our project, we will reconstruct the primary breast tumor and metastatic target organs such as liver, lung and bone (the most common clinical breast cancer metastasis organs). We will do so by creating small (" "Data-driven model inclusion in physics based mechanical models." "Frank Naets" "Mecha(tro)nic System Dynamics (LMSD)" "In recent years, data driven machine learning methods for modeling mechanical system dynamics have made huge leaps. However, these methods typically require large amounts of experimental data, which is often missing for practical applications, and do not exploit prior knowledge available from physics based modeling in mechanical systems. On the other hand several researchers have made first advancements in showing how e.g. neural networks (single layer perceptrons) can be exploited for describing constitutive material behavior in mechanics, without addressing the issue of how to train these models from experimental data.In this project we aim to develop a proof of concept on the integrated exploitation of mechanical finite element models with data driven constitutive laws. In particular we will develop a first scheme where nonlinear model order reduction is exploited to enable an efficient inverse analysis of the constitutive model parameters from experimental data." "Clinical risk prediction models based on multicenter data: methods for model development and validation" "Sabine Van Huffel" "Woman and Child, ESAT - STADIUS, Stadius Centre for Dynamical Systems, Signal Processing and Data Analytics" "Risk prediction models are developed to assist doctors in diagnosing patients, decision-making, counseling patients or providing a prognosis. To enhance the generalizability of risk models, researchers increasingly collect patient data in different settings and join forces in multicenter collaborations. The resulting datasets are clustered: patients from one center may have more similarities than patients from different centers, for example, due to regional population differences or local referral patterns. Consequently, the assumption of independence of observations, underlying the most often used statistical techniques to analyze the data (e.g., logistic regression), does not hold. This is mostly ignored in much of the current clinical prediction research. Research that relies on faulty assumptions may yield misleading results and lead to suboptimal improvements in patient care.To address this issue, I investigated the consequences of ignoring the assumption of independence and studied alternative techniques that acknowledge clustering throughout the process of planning a study, building a model and validating models in new data. I used mixed and random effects methods throughout the research as they allow to explicitly model differences between centers, and evaluated the proposed solutions with simulations and real clinical data. This dissertation covers sample size requirements, data collection and predictor selection, model fitting, and the validation of risk models in new data, focusing mainly on diagnostic models. The main case study is the development and validation of models for the pre-operative diagnosis of ovarian cancer, for which the multicenter dataset collected by the International Ovarian Tumor Analysis (IOTA) consortium is used.The results suggested that mixed effects logistic regression models offer center-specific predictions that have a better predictive performance in new patients than the predictions from standard logistic regression models. Although simulations showed that models were severely overfitted with only five events per variable, mixed effects models did not require more demanding sample size guidelines than standard logistic regression models. A case study on predictors of ovarian malignancy demonstrated that in multicenter data, measurements may vary systematically from one center to another, indicating potential threats to generalizability. These predictors could be detected using the residual intraclass correlation coefficient and may be excluded from risk models. In addition, a case study showed that, if statistical variable selection is used, mixed effects models are required in every step of the selection procedure to prevent incorrect inferences. Finally, case studies on risk models for ovarian cancer demonstrated that the predictive performance of risk models varied considerably between centers. This could be detected using meta-analytic models to analyze discrimination, calibration and clinical utility.In conclusion, taking into account differences between centers during the planning of prediction research, the development of a model and the validation of risk predictions in new patients offers insight in the heterogeneity and better predictions in local settings. Many methodological challenges remain, among which the inclusion of predictor-by-center interactions, the optimal application of mixed effects models in new centers, and the refinement of techniques to summarize clinical utility in multicenter data. Nonetheless, the findings in this dissertation imply that current clinical prediction research would benefit from adopting mixed and random effects techniques to fully employ the information that is available in multicenter data." "Advertisements showing non-ideal models and young women's and men's body image: development of a tripartite process model." "Steven Eggermont" "Leuven School for Mass Communication Research" "Beauty and fashion companies have been criticized for the persistent usage of extremely thin and muscular models in their advertisements (ads). Current literature has consistently shown that such idealized representations have detrimental effects on individual’s body image and general well-being. Society has therefore called for using more socially responsible advertising. Hence, advertisers have started to include “non-idealized models”, which are models who, according to social norms, do not comply to the conventional standards of beauty and are usually defined by having an average- or plus-size body size. Such non-idealized models are assumed to contribute to a more positive body image and lead to better advertising results. Yet, current research seems divided about whether such models actually do generate these expected positive outcomes.Therefore, the first goal of this PhD project was to map the effects of non-idealized models by systematically synthesizing prior literature. This systematic literature review is discussed in Chapter 1 and aimed to not only gain an overview of what is currently known about non-idealized models, but also mapped certain study characteristics that can provide insight in why current literature seems to be inconsistent. The results revealed that while for body image and well-being outcomes, most studies found either positive or null results, effects for advertising outcomes were more divided (N = 77 studies). These inconsistent findings can be explained by the large variety of labels and incomplete operationalizations that were identified within the review. The second goal of this PhD project was to assess under which conditions non-idealized models are most likely to have positive results. Three components are likely to determine the effectiveness of a non-idealized model: 1) message components – how are non-idealized models presented in an ad (i.e., framing and different types of non-idealized models), 2) source components – the corporate context that surrounds the ad (i.e., organization type), and 3) receiver components – the individual who is evaluating the ad (i.e., moderating psychosocial traits). These three components were investigated in different experiments, discussed in Chapter 2-4.Chapter 2 investigated the role of framing. Non-idealized model ads have been criticized for still being appearance-focused because they keep on framing the model as passive, only valued for her aesthetic, non-idealized appearance qualities (i.e., passive body framing). Alternative framing approaches emphasizing other qualities, such as body functionality (i.e., active body framing) or personality (i.e., subject framing), could be more successful in generating positive results. The experiment (N = 568 young women) discussed in Chapter 2 studied the effects of different framing approaches. Results showed that a passive object frame generated more appearance schemas than a subject frame and poorer ad outcomes as compared to an active object frame. The moderating role of thin-ideal internalization was also investigated, but no significant effects were found.Chapter 3 investigated the role of the type of organization that is distributing the non-idealized model ad. Using non-idealized models is seen as a corporate social responsibility (CSR) act. The effectiveness of a CSR act is dependent of the CSR fit, which means that results are most optimal when the social cause displayed in the ad, also fits the characteristics of the advertising organization itself. The two experiments in Chapter 3 investigated the effects of three types of CSR fit: 1) whether the function of the organization matches the idea of non-idealized models (profit vs. non-profit), 2) whether the image of the organization is known for supporting Body Positivity (BoPo) (poor vs. good reputation), and 3) whether the audience of the organization is open to non-idealized models (BoPo endorsement and gender). Results of both experiment 1 (N = 707 young women) and experiment 2 (N = 339 young women and men) revealed that a non-profit organization or a profit organization with a good reputation on BoPo resulted in better ad outcomes than a profit organization with a poor reputation. These effects were also mediated by external attribution, trustworthiness, and authenticity. No moderating effects of BoPo endorsement and gender were found.Chapter 4 investigated the role of different types of non-idealized models and focuses on men solely. For women, a non-idealized body generally refers to a larger body. Yet, for men, given the muscular body ideal, a non-idealized body can be both a larger as well as a skinnier body type. The intercultural experiment described in Chapter 4  (N = 363 young Flemish and Irish men) therefore investigated the effects of different body types (i.e., muscular, slim, larger, and a diversity condition with a combination of all types together) on men’s body image and advertising perceptions. Masculinity and country were also included as moderators. Findings show that the diversity condition showing a variety of body types generated the best effects for men’s body image.Overall, the current PhD project gave insights into the circumstances under which non-idealized models are most likely to have beneficial effects regarding body image and advertising perceptions. These results advance both body image and advertising research and theory and shed light on the inherent ambiguity that exists within non-idealized model ads." "Development of model order reduction techniques for mechatronic system models in inverse problems." "Wim Desmet" "Mecha(tro)nic System Dynamics (LMSD)" "Mechatronic models are developed in the design stage to describe and further improve the performance of systems. These models tend to cause difficulties with respect to numerical modelling and simulation due to (1) a large number of degrees of freedom (DOFs), (2) strongly coupled non-linear equations, (3) a wide frequency difference between the dominant behaviour of the different physical domains and (4) strongly non-linear parametric dependencies. To overcome the computational burden due to the large number of DOFs, the model should be reduced. The current state of the art in model order reduction is not entirely suited to deal in an efficient manner with the described characteristics. Therefore, the objective is to develop a non-linear, parametric model order reduction technique such that the reduced mechatronic model poses less problems in simulation. The research output should lead to parametric reduced models which describe complex systems but are computationally efficient and yet accurate. Models with these properties are particularly well suited for inverse analysis and design space explorations." "Merging satellite-derived and ground-based observations of spring phenology to select the best fitting and most parsimonious vegetation phenology model for global carbon cycle models." "Ivan Janssens" "Plant and Ecosystems (PLECO) - Ecology in a time of change" "Spring vegetation phenology determines the onset of the growing season. Changes in spring vegetation phenology alter the length of the growing season and thereby affect ecosystem productivity and regional and global carbon and energy balances. Satellite-derived vegetation indices have long been used as proxies for representing the status of terrestrial vegetation.However, the modeling of such large scale vegetation phenology dynamics is still a big challenge because the underlying mechanisms of vegetation phenology process are still unclear. To date, the performance of vegetation phenology models at global scale is rarely examined.Within this project, global-scale vegetation phenology models will be developed based on specieslevel models. Bayesian model comparisons will subsequently be conducted to select the most parsimonious vegetation phenology model for global carbon cycle models. In addition, remote sensing-based phenological dates will be compared to ground observations at species level to answer whether the satellite images capture the phenology dynamics observed in situ. This project also aims to explore the recent controversial debate on the amplitude of the advancement of spring phenology since the 1980s. The present study will make a step forward in the study of vegetation phenology and will have important implications for the ecological modeling community by suggesting the most optimal phenology model." "Robust and sparse methods to model mean and dispersion behavior in Generalized Linear Models." "Applied mathematics" "The Generalized Linear Model (GLM) is a very popular and flexible class of regression models that generalizes ordinary linear regression by allowing for example non-normal response variables. Logistic regression, which is widely used for binary classification, and Poisson regression, often used to model count data, both belong to this class. The parameters are typically estimated using maximum likelihood, but this very often leads to various problems when analyzing real data from practice. Firstly, outliers in the data may heavily influence classical methods, yielding unreliable results. Secondly, estimation and interpretability becomes very difficult or impossible when the number of variables becomes very high. Thirdly, real data often display a more complex dispersion behavior than expected under the GLM model. To solve these issues, sparse and robust estimation methods that model simultaneously the mean and the dispersion behavior in the context of GLMs will be developed. Their mathematical properties will be thoroughly investigated. The newly proposed methods should also be computationally efficient such that modern large datasets can be analyzed easily. Open-access user-friendly software will be provided." "Model estimation and selection for multiresolution graphical models." "Gerda Claeskens" "Operations Research and Statistics Research Group (ORSTAT) (main work address Leuven)" "The main goal of the project is to develop and validate methods to estimate networks at different levels or resolutions. One of the aims is to determine which level is most appropriate to estimate and interpret such network models and this will help researchers in the field to better understand the properties and the characteristics of the appropriate models that should be used to analyze the available data. The techniques I propose are oriented towards the estimation aspect (selecting a graphical object is inherently connected to selecting the nodes between which edges are placed and to the type of edges that one should place between nodes) as well as to a thorough and rigorous study of theoretical properties in a multiresolution framework. The multiresolution aspect relates to having collected data at different levels of coarseness and as such, one is interested in selecting an appropriate level of coarseness. Natural contexts where such situations can occur are, for example, financial applications, image denoising, gene expression data and functional magnetic resonance imaging (fMRI). In the analysis of brain connectivity from fMRI data, a scientist takes a series of measurements on brain regions which range from being very coarse (relatively large in size) to being very fine (relatively small in size). Once new sound methodologies are created, I proceed with proposing extensions with the aim to relax constraining assumptions or towards other classes of models." "Evaluating process model quality: do discovered process models only contain system behavior and nothing more?" "Benoit DEPAIRE" "Business Informatics" "Process mining concerns the discovery of process models based on observed process behavior. Over the last decades, many process discovery algorithms have been developed, each with their own strengths. To support further scientific progress in this domain, the community is in need of a strong evaluation framework for process discovery techniques. Currently, four building blocks for such a framework can be identified, i.e. a set of evaluation measures, an evaluation methodology, benchmark data sets and a programming environment to automate algorithm evaluation and comparison. The set of evaluation measures is the building block which has received most attention so far. The four most studied and applied quality dimensions are replay fitness, precision, generalization and simplicity. Until today, quantifying generalization, which measures the alignment of the discovered model with the true process, constitutes a persistent problem within process mining. The objective of this research proposal is therefore to improve a recently developed metric that aims at closing this research gap. This metric will estimate the likelihood that the discovered model produced the observed event log. In particular, the metric will allow both academia and practitioners to judge whether a model does not contain too much behavior, and thereby is suffering from a lack of realism."