In recent years, there has been a growing need for causal analysis in many modern stream applications such as web page click monitoring, patient health care monitoring, stock market prediction, electric grid monitoring, and network intrusion detection systems. The detection and prediction of causal relationships help in monitoring, planning, decision making, and prevention of unwanted consequences. An event stream is a continuous unbounded sequence of event instances. The availability of a large amount of continuous data along with high data throughput poses new challenges related to causal modeling over event streams, such as (1) the need for incremental causal inference for the unbounded data, (2) the need for fast causal inference for the high throughput data, and (3) the need for real-time prediction of effects from the events seen so far in the continuous event streams. This dissertation research addresses these three problems by focusing on utilizing temporal precedence information which is readily available in event streams: (1) an incremental causal model to update the causal network incrementally with the arrival of a new batch of events instead of storing the complete set of events seen so far and building the causal network from scratch with those stored events, (2) a fast causal model to speed up the causal network inference time, and (3) a real-time top-k predictive query processing mechanism to find the most probable k effects with the highest scores by proposing a run-time causal inference mechanism which addresses cyclic causal relationships. In this dissertation, the motivation, related work, proposed approaches, and the results are presented in each of the three problems.

No description available.

The purpose of this study was to propose an exploratory causal model that examines the influence of several antecedent variables on burnout. The antecedent variables included age, marital status, education, tenure, Type A personality, Jungian types, death anxiety, leadership style, job satisfaction, stress, coping efficacy, and marital satisfaction. The validity of the causal model was tested by using path analysis. Subjects were 100 male firefighters who completed self-report measures of the predictor variables. Instruments included the Jenkins Activity Survey, Myers- Briggs Type Indicator, Collett-Lester Attitudes Toward Death Scale, Leader Behavior Description Questionnaire, Job Descriptive Index, Perceived Job Stress, The Coping Inventory, Dyadic Adjustment Scale, and the Maslach Burnout Inventory. Perceived work stress made the only direct contribution to the variance in burnout. Direct paths were found to stress from job satisfaction, Type A personality, and single marital status. Job satisfaction was directly related to leadership (consideration) and the Jungian Introversion, Feeling, and Perceiving preferences. Direct paths were found to marital satisfaction from death anxiety, leadership (consideration), and leadership (structure). Leadership (consideration) was directly related to structure. From the above results, it can be concluded that perception of stress is an important factor in predicting burnout. Other factors are important contributors to stress and have indirect effects on burnout. Implications for the prevention and treatment of job burnout are discussed.

Modern DBMSes are designed to support many transactions running simultaneously. DBMS thrashing is indicated by the existence of a sharp drop in transaction throughput. The thrashing behavior in DBMSes is a serious concern to DBAs engaged in on-line transaction processing (OLTP) and on-line analytical processing (OLAP) systems, as well as to DBMS implementors developing technologies related to concurrency control. If thrashing is prevalent in a DBMS, thousands of transactions may be aborted, resulting in little progress in transaction throughput over time. From an engineering perspective, therefore, it is of critical importance to understand the factors of DBMS thrashing. However, understanding the origin of modern DBMSes' thrashing is challenging, due to many factors that may interact. The existing literature on thrashing exhibits the following weaknesses: (i) methodologies have been based on simulation and analytical studies, rather than on empirical analysis on real DBMSes, (ii) scant attention has been paid to the associations between factors, and (iii) studies have been restricted to one specific DBMS rather than across multiple DBMSes. This dissertation aims at better understanding the thrashing phenomenon across multiple DBMSes. We identify the underlying causes and propose a novel structural causal model to explicate the relationships between various factors contributing to DBMS thrashing. Our model derives a number of specific hypotheses to be subsequently tested across DBMSes, providing empirical support for this model as well as engineering implications for fundamental improvements in transaction processing. Our model also guides database researchers to refine this causal model, by looking into other unknown factors.

The Principal Stratification method estimates a causal intervention effect by taking account of subjects' differences in participation, adherence or compliance. The current Principal Stratification method has been mostly used in randomized intervention trials with randomization at a single (individual) level with subjects who were randomly assigned to either intervention or control condition. However, randomized intervention trials have been conducted at group level instead of individual level in many scientific fields. This is so called "two-level randomization", where randomization is conducted at a group (second) level, above an individual level but outcome is often observed at individual level within each group. The incorrect inferences may result from the causal modeling if one only considers the compliance from individual level, but ignores it or be determine it from group level for a two-level randomized trial. The Principal Stratification method thus needs to be further developed to address this issue. To extend application of the Principal Stratification method, this research developed a new methodology for causal inferences in two-level intervention trials which principal stratification can be formed by both group level and individual level compliance. Built on the original Principal Stratification method, the new method incorporates a range of alternative methods to assess causal effects on a population when data on exposure at the group level are incomplete or limited, and are data at individual level. We use the Gatekeeper Training Trial, as a motivating example as well as for illustration. This study is focused on how to examine the intervention causal effect for schools that varied by level of adoption of the intervention program (Early-adopter vs. Later-adopter). In our case, the traditional Exclusion Restriction Assumption for Principal Stratification method is no longer hold. The results show that the intervention had a stronger impact on Later-Adopter group than Early-Adopter group for all participated schools. These impacts were larger for later trained schools than earlier trained schools. The study also shows that the intervention has a different impact on middle and high schools.

De nouveaux traitements comme l’immunothérapie ont été proposés en oncologie. Ils sont basés sur les mécanismes de régulation du système immunitaire. Cependant tous les patients ne répondent pas à ces nouveaux traitements. Afin de pouvoir les identifier, on mesure l’association des marqueurs immunologiques exprimés à la réponse au traitement ainsi qu’à la toxicité à l’instaurationdu traitement et leur évolution sous traitement. En situation observationnelle, l’absence de tirage au sort empêche la comparabilité des groupes et l'effet mesuré est juste une mesure d'association. Les méthodes d’inférence causalepermettent dans certains cas, après avoir identifié les sources de biais de par la construction de diagrammes acycliques dirigés (DAG), d'atteindre l’interchangeabilité conditionnelle entre exposés et non exposés etpermettent l’estimation d’effets causaux. Dans les cas les plus simples où le nombre de variables est faible, il est possible de dessiner leDAG à partir d’expertise. Dans les situations où le nombre de variables explosent, des algorithmes d’apprentissage ont été proposés pour retrouver la structure de ces graphes. Néanmoins ces algorithmes font d’une part l’hypothèse qu’aucune information n’est connue et n’ont été développés que dans les cas où les covariables sont mesurés à un seul temps. L’objectif de cette thèse est donc de développer ces méthodes d’apprentissages de graphes à des données répétées, puis d’intégrer des connaissances a priori pour améliorer l’estimation de ceux-ci. Une fois les graphes appris les modèles causaux peuvent être appliqués sur les biomarkers immunologiques répétés pour détecter ceux qui sont associés à laréponse et/ou la toxicité. / In oncology, new treatments such as immunotherapy have been proposed, which are based on regulation of the immune system. However, not all treated patient have a long-term benefit of the treatment. To identify those patients who benefit most, we measured markers of the immune system expressed at treatment initiation and across time. In an observational study, the lack of randomization makes the groups not comparable and the effect measured is just an association. In this context, causal inference methods allow in some cases, after having identified all biases by constructing a directed acyclic graph (DAG), to get close to the case of conditional exchangeability between exposed and non-exposed subjects and thus estimating causal effects.In the most simple cases, where the number of variables is low, it is possible to draw the DAG with experts’ beliefs. Whereas in the situation where the number of variables rises, learning algorithms have been proposed in order to estimate the structure of the graphs. Nevertheless, these algorithms make the assumptions that any a priori information between the markers is known and have mainly been developed in the setting in which covariates are measured only once. The objective of this thesis is to develop learning methods of graphs for taking repeated measures into account, and reduce the space search by using a priori expert knowledge. Based on these graphs, we estimate causal effects of the repeated immune markers on treatment response and/or toxicity.

No description available.

A aplicação de Redes Bayesianas como modelo causal em Risco Operacional e extremamente atrativa do ponto de vista do gerenciamento dos riscos e do calculo do capital regulatorio do primeiro pilar do Novo Acordo da Basileia. Com as Redes e possível obter uma estimativa do VAR operacional utilizando-se não somente os dados históricos de perdas, mas também variáveis explicativas e conhecimento especialista através da possibilidade de inclusão de informações subjetivas. / The application of Bayesian Networks as causal model in Operational Risk is very attractive from the point of view of risk management and the calculation of regulatory capital under the first pillar of the New Basel Accord. It is possible to obtain with the networks an estimate of operational VAR based not only on the historical loss data but also in explanatory variables and expert knowledge through the possibility of inclusion of subjective information.

A aplicação de Redes Bayesianas como modelo causal em Risco Operacional e extremamente atrativa do ponto de vista do gerenciamento dos riscos e do calculo do capital regulatorio do primeiro pilar do Novo Acordo da Basileia. Com as Redes e possível obter uma estimativa do VAR operacional utilizando-se não somente os dados históricos de perdas, mas também variáveis explicativas e conhecimento especialista através da possibilidade de inclusão de informações subjetivas. / The application of Bayesian Networks as causal model in Operational Risk is very attractive from the point of view of risk management and the calculation of regulatory capital under the first pillar of the New Basel Accord. It is possible to obtain with the networks an estimate of operational VAR based not only on the historical loss data but also in explanatory variables and expert knowledge through the possibility of inclusion of subjective information.

<p dir="ltr">Recent advancements in representation learning have significantly enhanced the analysis of relational data across various domains, including social networks, bioinformatics, and recommendation systems. In general, these methods assume that the training and test datasets come from the same distribution, an assumption that often fails in real-world scenarios due to evolving data, privacy constraints, and limited resources. The task of out-of-distribution (OOD) extrapolation emerges when the distribution of test data differs from that of the training data, presenting a significant, yet unresolved challenge within the field. This dissertation focuses on developing robust representations for effective OOD extrapolation, specifically targeting relational data types like graphs and sets. For successful OOD extrapolation, it's essential to first acquire a representation that is adequately expressive for tasks within the distribution. In the first work, we introduce Set Twister, a permutation-invariant set representation that generalizes and enhances the theoretical expressiveness of DeepSets, a simple and widely used permutation-invariant representation for set data, allowing it to capture higher-order dependencies. We showcase its implementation simplicity and computational efficiency, as well as its competitive performances with more complex state-of-the-art graph representations in several graph node classification tasks. Secondly, we address OOD scenarios in graph classification and link prediction tasks, particularly when faced with varying graph sizes. Under causal model assumptions, we derive approximately invariant graph representations that improve extrapolation in OOD graph classification task. Furthermore, we provide the first theoretical study of the capability of graph neural networks for inductive OOD link prediction and present a novel representation model that produces structural pairwise embeddings, maintaining predictive accuracy for OOD link prediction as the test graph size increases. Finally, we investigate the impact of environmental data as a confounder between input and target variables, proposing a novel approach utilizing an auxiliary dataset to mitigate distribution shifts. This comprehensive study not only advances our understanding of representation learning in OOD contexts but also highlights potential pathways for future research in enhancing model robustness across diverse applications.</p>