Bridges are the most critical components of the transportation network. The functionality of bridges is important for hurricane aftermath recovery and emergency activities. However, past hurricane events revealed the potential susceptibility of these bridges under storm induced wave and surge loads. Coastal bridges traditionally were not designed to sustain hurricane induced wave and surge loads; and furthermore, no reliability assessment tool exists for bridges exposed to this hazard. However, such a tool is imperative for decision makers to evaluate the risk posed to the existing bridge inventory, and to decide on the retrofit measures and mitigation strategies. This dissertation offers a first attempt to quantify the structural vulnerability of bridges under coastal storms, offering a probabilistic framework, input tools, and application illustrations. To accomplish this goal, first an unbiased wave load model is developed based on the existing wave load models in the literature. The biased is removed from the load models through statistical analysis of the experimental test data. The developed wave load model is used to evaluate the response of coastal bridges employing single-physics domain Dynamic numerical models. Additionally, a high fidelity fluid-structure interaction model is developed to take into account the significant intricacies, such as turbulence, wave diffraction, and air entrapment, as well as material and geometric nonlinearities in structure. This numerical model provides insight on the influential parameters that affect the response of coastal bridges. Moreover, a Monte Carlo based Static Model methodology is developed to enable fast evaluation of the bridge deck unseating mode of failure. This methodology can be used for fast screening of vulnerable structures under hurricane induced wave and surge loads in a large bridge inventory. New statistical learning tools are used to develop fragility surfaces for coastal bridges vulnerable to storms. The performance of each of these tools is evaluated and compared. The statistical learning approaches are used to enable reliability assessment using the more rigorous finite element models such as the Dynamic and FSI Models which is important for improved confidence and retrofit assessment. Additionally, a new systematic method to evaluate the limit state capacity functions based on the post-event global performance of the bridge structure is developed. The application of the developed reliability models is illustrated by utilizing them for Houston/Galveston Bay area bridge inventory. The case study of Houston/Galveston Bay area reveals that more than 30% of bridges have a high probability of failure during an extreme hurricane scenario event. Two vulnerable bridge structures from the case study are selected to investigate the effect of different potential retrofit measures. Recommendations are made for the most appropriate retrofit measures that can prevent the deck unseating without significantly increasing the structural demands on other components.

When conducting a visual search task participants respond faster to targets embedded in a repeated array of visual distractors compared to targets embedded in a novel array, an effect referred to as contextual cueing. There are no reports of contextual cueing in audition, and generalizing this effect to the auditory domain would provide a new paradigm to investigate similarities, differences, and interactions in visual and auditory processing. In 4 experiments, participants identified a numerical target embedded in a sequence of alphabetic letter distractors. The training phase (Epochs 1, 2, and 3) of all experiments contained repeated sequences, and the testing phase (Epoch 4) contained novel sequences. Temporal contextual cueing was measured as slower response times in Epoch 4 than in Epoch 3. Repeated context was defined by the order of distractor identities and the rhythmic structure of the portion of the sequence immediately preceding the target digit, either together (Experiments 1 and 2) or separately (Experiments 3 and 4). An auditory temporal contextual cueing effect was obtained in Experiments 1, 2, and 4. This is the first report of an auditory temporal contextual cueing effect and, thus, it extends the contextual cueing effect to a new modality. This new experimental paradigm could be useful in furthering our understanding of fundamental auditory processes and could eventually be used to aid in diagnosing language deficits.

Spatial neglect has mainly been described through its spatial deficits (such as attentional bias, disengagement deficit or exploratory motor behavior), but numerous other studies have reported non-spatial impairments in patients suffering from this disorder. In the present thesis, non-spatial deficits in neglect are hypothesized to form a core impairment, which can be summarized as a difficulty to learn and benefit from regularities in the environment. The different studies conducted and reported in the present thesis have converged to support this hypothesis that neglect patients have difficulty to interact with environmental statistics. The two first studies, which tested the visual modality (Chapters 2 and 3), have demonstrated that neglect patients have difficulties to become faster to respond to targets that appear successively at the same location (position priming). This difficulty is also more generic, as neglect patients do not learn that some things occur more often than others, such as for example that a target has a high probability to be repeated at a specific region. Those two studies have shown that neglect patients are impaired in position priming and statistical learning, which corresponds to difficulties benefiting from regularities presented in the visual domain. This difficulty may be explained by patients’ impairment in working memory or temporal processing. Several studies have reported the implication of those two mechanism in statistical learning: if patients tend to underestimate the time that a target is presented on the screen and have difficulties keeping in memory its precedent location, this translates into a difficulty to benefit from the repeats of the target position as well as a difficulty to benefit from transition probability. In order to verify if priming and statistical learning impairments were specific to the visual modality or if neglect patients present a multimodal difficulty to learn the transition probability in general, brain damaged patients were tested in the auditory domain (Chapter 5), with a paradigm that has shown statistical learning in infants. This study confirmed that for the auditory modality too, brain damaged patients are impaired in statistical learning. The different results of the studies reported in Chapters 2, 3, 4 and 5 converge to support the hypothesis that spatial neglect patients have difficulties benefiting from regularities of their environment. Nevertheless, this impairment is not irreversible, as it was demonstrated by a chronic neglect patient who was trained with three sessions distributed over three days (Chapter 2). Although having similar results to the other patients for the first session, this patients’ performance improved over the sessions to show a faster reaction time for the targets presented on the high probability region (his contralesional side). Therefore, priming and statistical learning investigated in this thesis are worth exploring further for their potential outcome in the rehabilitation domain.

This paper bridges the gap between variable selection methods (e.g., Pearson coefficients, KS test) and dimensionality reductionalgorithms (e.g., PCA, LDA). Variable selection algorithms encounter difficulties dealing with highly correlated data,since many features are similar in quality. Dimensionality reduction algorithms tend to combine all variables and cannotselect a subset of significant variables.Our approach combines both methodologies by applying variable selection followed by dimensionality reduction. Thiscombination makes sense only when using the same utility function in both stages, which we do. The resulting algorithmbenefits from complex features as variable selection algorithms do, and at the same time enjoys the benefits of dimensionalityreduction.1

Statistical learning refers to the ability to track regular patterns in sensory input from ambient environments. This learning mechanism can exploit a wide range of statistical structures (e.g., frequency, distribution, and co-occurrence probability). Given its regularities and hierarchical structures, language is essentially a pattern-based system and therefore researchers have argued that statistical learning is fundamental to language acquisition (e.g., Saffran, 2003). Indeed, young infants and adults can find words in artificial languages by tracking syllable co-occurrence probabilities and extracting words on that basis (e.g., Saffran. Aslin & Newport, 1996a). However, prior studies have mainly focused on whether learners can statistically segment words from a single language; whether learners can segment words from two artificial languages remains largely unknown. Given that the majority of the global population is bilingual (Grosjean, 2010), it is necessary to study whether learners can make use of the statistical learning mechanism to segment words from two language inputs, which is the focus of this thesis. I examined adult and infant learners to answer three questions: (i) Can learners make use of French and English phonetic cues within a single individual’s speech to segment words successfully from two languages?; 2) Do bilinguals outperform monolinguals?; and 3) Do specific factors, such as cognitive ability or bilingual experience, underlie any potential bilingual advantage in word segmentation across two languages? In Study 1, adult learners generally could make use of French and English phonetic cues to segment words from two overlapping artificial languages. Importantly, simultaneous bilinguals who learned French and English since birth segmented more correct words in comparison to monolinguals, multilinguals, and sequential French-English bilinguals. Early bilingual experience may lead learners to be more flexible when processing information in new environments and/or they are more sensitive to subtle cues that mark the changes of language inputs. Further, individuals’ cognitive abilities were not related to learners’ segmentation performance, suggesting that the observed simultaneous bilingual segmentation advantage is not related any bilingual cognitive advantages (Bialystok, Craik, & Luk, 2012). In Study 2, I tested 9.5-month-olds, who are currently discovering words in their natural environment, in an infant version of the adult task. Surprisingly, monolingual, but not bilingual, infants successfully used French and English phonetic cues to segment words from two languages. The observed difference in segmentation may be related to how infant process native and non-native phonetic cues, as the French phonetic cues are non-native to monolingual infants but are native to bilingual infants. Finally, the observed difference in segmentation ability was again not driven by cognitive skills. In sum, current thesis provides evidence that both adults and infants can make use of phonetic cues to statistically segment words from two languages. The implications of why early bilingualism plays a role in determining learners’ segmentation ability are discussed.

Machine learning is a buzz word that has inundated popular culture in the last few years. This is a term for a computer method that can automatically learn and improve from data instead of being explicitly programmed at every step. Investigations regarding the best way to create and use these methods are prevalent in research. Machine learning models can be difficult to create because models need to be tuned. This dissertation explores the characteristics of tuning three popular machine learning models and finds a way to automatically select a set of tuning parameters. This information was used to create an R software package called EZtune that can be used to automatically tune three widely used machine learning algorithms: support vector machines, gradient boosting machines, and adaboost. The second portion of this dissertation investigates the implementation of machine learning methods in finding locations along a genome that are associated with a trait. The performance of methods that have been commonly used for these types of studies, and some that have not been commonly used, are assessed using simulated data. The affect of the strength of the relationship between the genetic code and the trait is of particular interest. It was found that the strength of this relationship was the most important characteristic in the efficacy of each method.

Model-based clustering is a probabilistic approach that views each cluster as a component in an appropriate mixture model. The Gaussian mixture model is one of the most widely used model-based methods. However, this model tends to perform poorly when clustering high-dimensional data due to the over-parametrized solutions that arise in high-dimensional spaces. This work instead considers the approach of combining dimension reduction techniques with clustering via a mixture of generalized hyperbolic distributions. The dimension reduction techniques, principal component analysis and factor analysis along with their extensions were reviewed. Then the aforementioned dimension reduction techniques were individually paired with the mixture of generalized hyperbolic distributions in order to demonstrate the clustering performance achieved under each method using both simulated and real data sets. For a majority of the data sets, the clustering method utilizing principal component analysis exhibited better classi cation results compared to the clustering method based on the extending the factor analysis model. / Thesis / Master of Science (MSc)

This thesis studies the problem of frequentist model averaging over a set of multiple $\epsilon$-support vector regression (SVR) models, where the support vector machine (SVM) algorithm was extended to function estimation involving continuous targets, instead of categorical ones. By assigning weights to a set of candidate models instead of selecting the least misspecified one, model averaging presents a strong alternative to model selection for tackling model uncertainty. Not only do we describe the construction of smoothed BIC/AIC model averaging weights, but we also propose a Mallows model averaging procedure which selects model weights by minimizing Mallows' criterion. We conduct two studies where the set of candidate models can either include or not include the true model by making use of simulated random samples obtained from different data-generating processes of analytic form. In terms of mean squared error, we demonstrate that our proposed method outperforms other model averaging and model selection methods that were tested, and the gain is more substantial for smaller sample sizes with larger signal-to-noise ratios. / Thesis / Master of Science (MSc)

The human desire to explain the world is the driving force behind our species' rich history of scientific and technological advancement. The ability of successive generations to build cumulatively on the scientific progress made by their ancestors rests on the ability of individual minds to rapidly assimilate the explanatory models developed by those who came before. But is this explanatory, model-based way of thinking limited to deliberate, conscious cognition, with the larger, unconscious portion of the workings of the mind dependent on simpler mechanisms of association and prediction, or is explanation a more fundamental drive? In this dissertation I explore theoretical, empirical and computational attempts to shed some light on this question. I first present a number of theoretical advantages that model-based learning has over its associative counterparts. I focus particularly on the inferential phenomenon of \emph{explaining away}, which is difficult to account for in a model-free system of learning. Next I review some recent empirical literature which helps to establish just what mechanisms of learning are available to human infants and adults, including a number of findings that suggest that there is more to learning than mere prediction. Among these are a number of experiments suggesting that explaining away occurs in a variety of cognitive domains. Having set the stage, I report a new set of experiments, one with infants and two with adults, along with a related computational model, which provide further evidence for unconscious explaining away, and hence for some for of model-based inference, in the domain of abstract, relational pattern-learning. In particular, I find that when learners are presented with a novel environment of tone sequences, the structure of their initial experience with that environment, and implicitly the model of the environment which best accounts for that experience, influences what kinds of abstract structure can easily be learned later. If indeed learners are able to construct explanatory models of particular domains of experience which are then used to learn the details of each domain, it may undermine claims by some philosophers and cognitive scientists that asymmetries in learning across domains constitutes evidence for an innately modular organization of the mind.

How do infants segment objects from the complex visual environment? Investigations of figure-ground perception have been dominated by studies assessing infants' sensitivity to depth and figure cues; few studies have assessed what information infants' use to perceive figures as separate from grounds. Research examining word segmentation suggests that statistical learning might aid segmentation in visual scenes. Despite the numerous studies investigating figure-ground segmentation, none have investigated the role of spatial transitional probabilities as a means for segmentation. To examine this question, we used a habituation/familiarity-preference procedure to assess whether background variability enables 5.5-month-old infants' perception of figures as separate from the background. The results of Experiments 1 and 2 indicated that statistical learning extends to scene segmentation, scene contexts allowed extraction of statistical distribution. Experiment 3 demonstrated that matching the configuration of visual arrays during training and test is essential; mismatched stimuli impede the measurement of segmentation.