Neurological disorders present a significant challenge in global health. With the increasing availability of imaging datasets and the development of precise machine learning models, early and accurate diagnosis of neurological conditions is a promising and active area of research. However, several characteristic factors in neurology domains, such as heterogeneous imaging, inaccurate labels, or limited data, act as bottlenecks in using deep learning on clinical neuroimaging. Given these circumstances, this dissertation attempts to provide a guideline, proposing several methods and showcasing successful implementations in broad neurological conditions, including epilepsy and neurodegeneration. Methodologically, a particular focus is on comparing a two-dimensional approach as opposed to three-dimensional neural networks. In most clinical domains of neurological disorders, data are scarce and signals are weak, discouraging the use of 3D representation of raw scan data. This dissertation first demonstrates competitive performances with 2D models in tuber segmentation and AD comorbidity detection. Second, the potentials of ensemble learning are explored, further justifying the use of 2D models in the identification of neurodegeneration. Lastly, CleanNeuro is introduced in the context of 2D classification, a novel algorithm for denoising the datasets prior to training. CleanNeuro, on top of 2D classification and ensemble learning, demonstrates the feasibility of accurately classifying patients with comorbid AD and cerebral amyloid angiopathy from AD controls. Methods presented in this dissertation may serve as exemplars in the study of neurological disorders using deep learning and clinical neuroimaging. Clinically, this dissertation contributes to improving automated diagnosis and identification of regional vulnerabilities of several neurological disorders on clinical neuroimaging using deep learning. First, the classification of patients with Alzheimer’s disease from cognitively normal group demonstrates the potentials of using positron emission tomography with tau tracers as a competitive biomarker for precision medicine. Second, the segmentation of tubers in patients with tuberous sclerosis complex proves a successful 2D modeling approach in quantifying neurological burden of a rare yet deadly disease. Third, the detection of comorbid pathologies from patients with Alzheimer’s disease is analyzed and discussed in depth. Based on prior findings that comorbidities of Alzheimer’s disease affect the brain structure in a distinctive pattern, this dissertation proves for the first time the effectiveness of using deep learning on the accurate identification of comorbid pathology in vivo. Leveraging postmortem neuropathology as ground truth labels on top of the proposed methods records competitive performances in comorbidity prediction. Notably, this dissertation discovers that structural magnetic resonance imaging is a reliable biomarker in differentiating the comorbid cereberal amyloid angiopathy from Alzheimer’s disease patients. The dissertation discusses experimental findings on a wide range of neurological disorders, including tuberous sclerosis complex, dementia, and epilepsy. These results contribute to better decision-making on building neural network models for understanding and managing neurological diseases. With the thorough exploration, the dissertation may provide valuable insights that can push forward research in clinical neurology.

A search is presented for hadronic signatures of a strongly-coupled hidden dark sector, accessed via resonant production of a ?′ mediator. The analysis uses 139 fb-1 of proton-proton collision data collected by the ATLAS experiment during Run 2 of the LHC. The ?′ mediator decays to two dark quarks, which each hadronize and decay to showers containing both dark and Standard Model particles; these showers are termed “semi-visible” jets. The final state consists of missing energy aligned with one of the jets, a topology that is ignored by most dark matter searches. A supervised machine learning method is used to select these dark showers and reject the dominant background of mis-measured multijet events. A complementary semi-supervised anomaly detection approach introduces broad sensitivity to a variety of strongly coupled dark matter models. A resonance search is performed by fitting the transverse mass spectrum with a polynomial background estimation function. Results are presented as limits on the effective cross section of the Z', parameterized by the fraction of invisible particles in the decay and the Z' mass. No structure in the transverse mass spectrum compatible with the signal hypothesis is observed. Z' mediator masses from ranging from 2.0 TeV to 3.5 TeV are excluded at the 95% confidence level.

Robots and other systems using deep learning are increasingly common. It is essential to accurately keep track of what they have done and determine if they are operating as we intend. In this dissertation, I introduce several approaches to enable better monitoring and understanding of these systems. I propose and demonstrate robot action summarization and question answering in natural language. The first step in understanding and controlling robots is knowing what they are doing. Much research has been done on training robots to learn to follow natural language instructions; robot action summarization is a complement to this research, enabling robots to report back succinctly and accurately what they have done after completing a task. I demonstrate that such summaries can be generated from multimodal inputs using recurrent and transformer networks. I then investigate question answering about robot action episodes using a dataset of questions and answers I introduce. I show that learning to answer questions can help a model summarize by enabling it to learn about some objects solely during question answering and then transferring that representational knowledge to summarization. If robots are to summarize and answer questions about their past actions, it will be necessary for them to store and recall episodes of action. I introduce a technique to form compact memory representations which can be used for these tasks, as well as for guiding choices about actions which should be taken during an action sequence. In addition to helping users keep track of what robots or other machine learning systems are doing, such artificial episodic memory representations could also pose some undesirable risks. I therefore propose a set of principles to guide the safe development of artificial episodic memory. Finally, I introduce a method to learn to predict a neural network’s accuracy on particular inputs by training a second network to examine the outputs from its intermediate, hidden layers or its final outputs.

The rapid growth of population and the permanent increase in the number of vehicles engender several issues in transportation systems, which in turn call for an intelligent and cost-effective approach to resolve the problems in an efficient manner. A cost-effective approach for improving and optimizing transportation-related problems is to unlock hidden knowledge in ever-increasing spatiotemporal and crowdsourced information collected from various sources such as mobile phone sensors (e.g., GPS sensors) and social media networks (e.g., Twitter). Data mining and machine learning techniques are the major tools for analyzing the collected data and extracting useful knowledge on traffic conditions and mobility behaviors. Deep learning is an advanced branch of machine learning that has enjoyed a lot of success in computer vision and natural language processing fields in recent years. However, deep learning techniques have been applied to only a small number of transportation applications such as traffic flow and speed prediction. Accordingly, my main objective in this dissertation is to develop state-of-the-art deep learning architectures for resolving the transport-related applications that have not been treated by deep learning architectures in much detail, including (1) travel mode detection, (2) vehicle classification, and (3) traffic information system. To this end, an efficient representation for spatiotemporal and crowdsourced data (e.g., GPS trajectories) is also required to be designed in such a way that not only be adaptable with deep learning architectures but also contains efficient information for solving the task-at-hand. Furthermore, since the good performance of a deep learning algorithm is primarily contingent on access to a large volume of training samples, efficient data collection and labeling strategies are developed for different data types and applications. Finally, the performance of the proposed representations and models are evaluated by comparing to several state-of-the-art techniques in literature. The experimental results clearly and consistently demonstrate the superiority of the proposed deep-learning based framework for each application. / PHD / The rapid growth of population and the permanent increase in the number of vehicles engender several issues in transportation systems, which in turn call for an intelligent and cost-effective approach to resolve the problems in an efficient manner. Furthermore, the recent advances in positioning tools (e.g., GPS sensors) and ever-popularity of social media networks have enabled generation of massive spatiotemporal and crowdsourced data. This dissertation aims to leverage the advances in artificial intelligence so as to unlock the rick knowledge in the recorded data and in turn, optimizing the transportation systems in a cost-effective way. In particular, this dissertation seeks for proposing end-to-end frameworks based on deep learning models, as an advanced branch of artificial intelligence, as well as spatiotemporal and crowdsourced datasets (e.g., GPS trajectory and social media) for improving three transportation problems. (1) Travel Mode Detection, which is defined as identifying users’ transportation mode(s) (e.g., walk, bike, bus, car, and train) when traveling around the traffic network. (2) Vehicle Classification, which is defined as identifying the vehicle’s type (e.g., passenger car and truck) while moving in a traffic network. (3) traffic information system based on social media networks, which is defined as detecting traffic events (e.g., crash) and capturing traffic information (e.g., traffic congestion) on a real-time basis from users’ tweets. The experimental results clearly and consistently demonstrate the superiority of the proposed deep-learning based framework for each application.

Active feedback control in magnetic confinement fusion devices is desirable to mitigate plasma instabilities and enable robust operation. Among various diagnostics, optical high-speed cameras provide a powerful, non-invasive diagnostic and can be suitable for these applications. This thesis reports the first application of high-speed imaging videography and deep learning as real-time diagnostics of rotating MHD modes in a tokamak device. The developed system uses a convolutional neural network (CNN) to predict the amplitudes of the ?=1 sine and cosine mode components using solely optical measurements acquired from one or more cameras. Using the newly assembled high-speed camera diagnostics on the High Beta Tokamak – Extended Pulse (HBT-EP) device, an experimental dataset consisting of camera frame images and magnetic-based mode measurements was assembled and used to develop the mode-tracking CNN model. The optimized models outperformed other tested conventional algorithms given identical image inputs. A prototype controller based on a field-programmable gate array (FPGA) hardware was developed to perform real-time mode tracking using the high-speed camera diagnostic with the mode-tracking CNN model. In this system, a trained model was directly implemented in the firmware of an FPGA device onboard the frame grabber hardware of the camera’s data readout system. Adjusting the model size and its implementation-related parameters allowed achieving an optimal trade-off between a model’s prediction accuracy, its FPGA resource utilization and inference speed. Through fine-tuning these parameters, the final implementation satisfied all of the design constraints, achieving a total trigger-to-output latency of 17.6 ?s and a throughput of up to 120 kfps. These results are on-par with the existing GPU-based control system using magnetic sensor diagnostic, indicating that the camera-based controller will be capable to perform active feedback control of MHD modes on HBT-EP.

Chronic obstructive pulmonary disease (COPD) and emphysema together are one of the leading causes of death in the United States and worldwide; meanwhile, breast cancer has the highest incidence and second-highest mortality burden of all cancers in women. Imaging markers relevant to each of these conditions are readily identifiable on chest computed tomography (CT): (1) visually-appreciable variants in airway tree structure exist which are associated with increased odds for development of COPD; (2) CT emphysema subtypes (CTES), based on lung texture and spatial features, have been identified by unsupervised clustering and correlate with functional measures and clinical outcomes; (3) dysanapsis, or the ratio of airway caliber to lung volume, is the strongest known predictor of COPD risk, and (4) breast density (i.e., the extent of fibroglandular tissue within the breast) is strongly associated with breast cancer risk. Machine- and deep-learning frameworks present an opportunity to address unmet needs in each of these directions, leveraging the data from large CT cohorts. Application of unsupervised learning approaches serves to discover new, image-based phenotypes. While topologic and geometric variation in the structure of the CT-resolved airway tree are well-described, tree- structural subtypes are not fully characterized. Similarly, while the clinical correlates of CTES have been described in large cohort studies, the association of CTES with structural and functional measures of the lung parenchyma are only partially described, and the time-dependent evolution of emphysematous lung texture has not been studied. Supervised approaches are required to automate CT image assessment, or to estimate CT- based measures from incomplete input data. While dysanapsis can be directly quantified on full- lung CT, the lungs are often only partially imaged in large CT datasets; total lung volume must then be regressed from the observed partial image. Breast density grades, meanwhile, are generally visually assessed, which is laborious to perform at scale. Moreover, current automated methods rely on segmentation followed by intensity thresholding, excluding higher-order features which may contribute to the radiologist assessment. In this thesis, we present a series of machine-learning methods which address each of these gaps in the field, using CT scans from the Multi-Ethnic Study of Atherosclerosis (MESA), the SubPopulations and InteRmediate Outcome Measures in COPD (SPIROMICS) Study, and an institutional chest CT dataset acquired at Columbia University Irving Medical Center. First, we design a novel graph-based clustering framework for identifying tree-structure subtypes in Billera-Holmes-Vogtmann (BHV) tree-space, using the airway trees segmented from the full-lung CT scans of MESA Lung Exam 5. We characterize the behavior of our clustering algorithm on a synthetic dataset, describe the geometric and topological variation across tree-structure clusters, and demonstrate the algorithm’s robustness to perturbation of the input dataset and graph tuning parameter. Second, in MESA Lung Exam 5 CT scans, we quantify the loss of small-diameter airway and pulmonary vessel branches within CTES-labeled lung tissue, demonstrating that depletion of these structures is concentrated within CTES regions, and that the magnitude of this effect is CTES-specific. In a sample of 278 SPIROMICS Visit 1 participants, we find that CTES demonstrate distinct patterns of gas trapping and functional small airways disease (fSAD) on expiratory CT imaging. In the CT scans of SPIROMICS participants imaged at Visit 1 and Visit 5, we update the CTES clustering pipeline to identify longitudinal emphysema patterns (LEPs), which refine CTES by defining subphenotypes informative of time-dependent texture change. Third, we develop a multi-view convolutional neural network (CNN) model to estimate total lung volume (TLV) from cardiac CT scans and lung masks in MESA Lung Exam 5. We demonstrate that our model outperforms regression on imaged lung volume, and is robust to same- day repeated imaging and longitudinal follow-up within MESA. Our model is directly applicable to multiple large-scale cohorts containing cardiac CT and totaling over ten thousand participants. Finally, we design a 3-D CNN model for end-to-end automated breast density assessment on chest CT, trained and evaluated on an institutional chest CT dataset of patients imaged at Columbia University Irving Medical Center. We incorporate ordinal regression frameworks for density grade prediction which outperform binary or multi-class classification objectives, and we demonstrate that model performance on identifying high breast density is comparable to the inter-rater reliability of expert radiologists on this task.

This dissertation comprises several studies addressing supervised learning problems where the supervision is imperfect. Firstly, we investigate the margin conditions in active learning. Active learning is characterized by its special mechanism where the learner can sample freely over the feature space and exploit mostly the limited labeling budget by querying the most informative labels. Our primary focus is to discern critical conditions under which certain active learning algorithms can outperform the optimal passive learning minimax rate. Within a non-parametric multi-class classification framework,our results reveal that the uniqueness of Bayes labels across the feature space serves as the pivotal determinant for the superiority of active learning over passive learning. Secondly, we study the estimation of central mean subspace (CMS), and its application in transfer learning. We show that a fast parametric convergence rate is achievable via estimating the expected smoothed gradient outer product, for a general class of covariate distribution that admits Gaussian or heavier distributions. When the link function is a polynomial with a degree of at most r and the covariates follow the standard Gaussian, we show that the prefactor depends on the ambient dimension d as d^r. Furthermore, we show that under a transfer learning setting, an oracle rate of prediction error as if the CMS is known is achievable, when the source training data is abundant. Finally, we present an innovative application involving the utilization of weak (noisy) labels for addressing an Individual Tree Crown (ITC) segmentation challenge. Here, the objective is to delineate individual tree crowns within a 3D LiDAR scan of tropical forests, with only 2D noisy manual delineations of crowns on RGB images available as a source of weak supervision. We propose a refinement algorithm designed to enhance the performance of existing unsupervised learning methodologies for the ITC segmentation problem.

Trajectory prediction aims to forecast future trajectories of agents (such as vehicles, pedestrians) based on their historical values. It is a fundamental step for advancing transportation management and control, directly impacting the safety and efficiency of modern transportation systems. In this research domain, deep learning-based methods have been widely adopted, achieving impressive performance. However, these methods have several drawbacks. First, they require substantial amounts of data. Second, they are prone to randomness inherent in real-world data. Third, complex interactions among transportation agents impose high demands on deep learning models. This dissertation seeks to address these challenges through physics-informed deep learning (PIDL), a promising approach that integrates physics-based prior knowledge into data-driven models. The dissertation is organized into three parts, focusing on different aspects of applying PIDL for trajectory prediction. First, we formulate the problem of single-agent trajectory prediction using PIDL. Second, we enhance PIDL by incorporating uncertainty quantification, accounting for uncertainties in both data and model parameters, and predicting future trajectories with confidence intervals. Third, we extend the single-agent trajectory prediction problem to a multi-agent setting, employing graph neural networks to model complex spatial interactions and NeuralODE to capture long-term dependencies. Through evaluations on both numerical and real-world datasets, our proposed methods demonstrate improved performance compared to state-of-the-art approaches. Moreover, leveraging physics-based prior knowledge makes our methods particularly robust in scenarios where deep learning models struggle, such as data-scarce environments and long-term predictions.

Many companies have a large collection of different model variants and results. Hexagon's (formerly MSC Software) software Odyssee helps to find out what information is contained in this data. New calculations can sometimes be avoided because the results for new parameter combinations can be predicted from the existing calculations. This is particularly interesting for non-linear or large models with long run times. The software also helps when setting up new DOEs and offers a variety of options for statistical displays. In the lecture, the number-based and image-based methods are compared. / Viele Firmen können auf eine große Sammlung vorhandener Rechnungen für verschiedene Modellvarianten zurückgreifen. Die Software Odyssee von Hexagon (früher MSC Software) hilft herauszufinden, welche Informationen in diesen Daten stecken. Neue Rechnungen kann man sich teilweise ersparen, weil die Ergebnisse für neue Parameterkombinationen aus den vorhandenen Rechnungen vorhergesagt werden können. Dies ist besonders interessant für nichtlineare oder große Modelle mit langer Rechenzeit. Die Software hilft auch beim Aufsetzen neuer DOEs und bietet vielfältige Möglichkeiten für statistische Darstellungen. In dem Vortrag werden die zahlenbasierte und bildbasierte Methode gegenübergestellt.

Submitted by Johnny Rodrigues (johnnyrodrigues@ufcg.edu.br) on 2018-05-01T14:25:43Z No. of bitstreams: 1 ADALBERTO GOMES TEIXEIRA JÚNIOR - DISSERTAÇÃO PPGCC 2015..pdf: 2611686 bytes, checksum: 6b9c4a2efc3946611ad0263328434bd1 (MD5) / Made available in DSpace on 2018-05-01T14:25:43Z (GMT). No. of bitstreams: 1 ADALBERTO GOMES TEIXEIRA JÚNIOR - DISSERTAÇÃO PPGCC 2015..pdf: 2611686 bytes, checksum: 6b9c4a2efc3946611ad0263328434bd1 (MD5) Previous issue date: 2015-07 / Capes / O som gerado por motores em funcionamento contém informações sobre seu estado e condições, tornando-se uma fonte importante para a avaliação de seu funcionamento sem a necessidade de intervenção no equipamento. A análise do estado do equipamento muitas vezes é realizada por diagnóstico humano, a partir da experiência vivenciada no ambiente ruidoso de operação. Como o funcionamento dos motores é regido por um processo periódico, o sinal de áudio gerado segue um padrão bem definido, possibilitando, assim, a avaliação de seu estado de funcionamento por meio desse sinal. Dentro deste contexto, a pesquisa ora descrita trata da modelagem do sinal acústico gerado por motores em usinas termoelétricas, aplicando técnicas de processamento digital de sinais e inteligência artificial, com o intuito de auxiliar o diagnóstico de falhas, minimizando a presença humana no ambiente de uma sala de motores. A técnica utilizada baseia-se no estudo do funcionamento dos equipamentos e dos sinais acústicos por eles gerados por esses, para a extração de características representativas do sinal, em diferentes domínios, combinadas a métodos de aprendizagem de máquinas para a construção de um multiclassificador, responsável pela avaliação do estado de funcionamento desses motores. Para a avaliação da eficácia do método proposto, foram utilizados sinais extraídos de motores da Usina Termoelétrica Borborema Energética S.A., no âmbito do projeto REPARAI (REPair over AiR using Artificial Intelligence, código ANEEL PD6471-0002/2012). Ao final do estudo, o método proposto demonstrou acurácia próxima a 100%. A abordagem proposta caracterizou-se, portanto, como eficiente para o diagnóstico de falhas, principalmente por não ser um método invasivo, não exigindo, portanto, o contato direto do avaliador humano com o motor em funcionamento. / The sound generated by an engine during operation contains information about its conditions, becoming an important source of information to evaluate its status without requiring intervention in equipment. The fault diagnosis of the engine usually is performed by a human, based on his experience in a noisy environment. As the operation of the engine is a periodic procedure, the generated signal follows a well-defined pattern, allowing the evaluation of its operating conditions. On this context, this research deals with modeling the acoustic signal generated by engines in power plants, using techniques from digital signal processing and artificial intelligence, with the purpose of assisting the fault diagnosis, minimizing the human presence at the engine room. The technique applied is based on the study of engines operation and the acoustic signal generated by them, extracting signal representative characteristics in different domains, combined with machine learning methods, to build a multiclassifier to evaluate the engines status. Signals extracted from engines of Borborema Energética S.A. power plant, during the REPARAI Project (REPair over AiR using Artificial Intelligence), ANEEL PD-6471-0002/2012, were used in the experiments. In this research, the method proposed has demonstrated an accuracy rate of nearly 100%. The approach has proved itself to be efficient to fault diagnosis, mainly by not being an invasive method and not requiring human direct contact with the engine.