This dissertation presents novel solutions for improving the generalization capabilities of deep learning based computer vision models. Neural networks are known to suffer a large drop in performance when tested on samples from a different distribution than the one on which they were trained. The proposed solutions, based on latent space geometry and meta-learning, address this issue by improving the robustness of these models to distribution shifts. Through the use of geometrical alignment, state-of-the-art domain adaptation and source-free test-time adaptation strategies are developed. Additionally, geometrical alignment can allow classifiers to be progressively adapted to new, unseen test domains without requiring retraining of the feature extractors. The dissertation also presents algorithms for enabling in-the-wild generalization without needing access to any samples from the target domain. Other causes of poor generalization, such as data scarcity in critical applications and training data with high levels of noise and variance, are also explored. To address data scarcity in fine-grained computer vision tasks such as object detection, novel context-aware augmentations are suggested. While the first four chapters focus on general-purpose computer vision models, strategies are also developed to improve robustness in specific applications. The efficiency of training autonomous agents for visual navigation is improved by incorporating semantic knowledge, and the integration of domain experts' knowledge allows for the realization of a low-cost, minimally invasive generalizable automated rehabilitation system. Lastly, new tools for explainability and model introspection using counter-factual explainers trained through interval-based uncertainty calibration objectives are presented.

As people begin to live longer and the population shifts to having more olderadults on Earth than young children, radical solutions will be needed to ease the burden on society. It will be essential to develop technology that can age with the individual. One solution is to keep older adults in their homes longer through smart home and smart living technology, allowing them to age in place. People have many choices when choosing where to age in place, including their own homes, assisted living facilities, nursing homes, or family members. No matter where people choose to age, they may face isolation and financial hardships. It is crucial to keep finances in mind when developing Smart Home technology. Smart home technologies seek to allow individuals to stay inside their homes for as long as possible, yet little work looks at how we can use technology in different life stages. Robots are poised to impact society and ease burns at home and in the workforce. Special attention has been given to social robots to ease isolation. As social robots become accepted into society, researchers need to understand how these robots should mimic natural conversation. My work attempts to answer this question within social robotics by investigating how to make conversational robots natural and reciprocal. I investigated this through a 2x2 Wizard of Oz between-subjects user study. The study lasted four months, testing four different levels of interactivity with the robot. None of the levels were significantly different from the others, an unexpected result. I then investigated the robot’s personality, the participant’s trust, and the participant’s acceptance of the robot and how that influenced the study.

Beta-Amyloid(Aβ) plaques and tau protein tangles in the brain are now widely recognized as the defining hallmarks of Alzheimer’s disease (AD), followed by structural atrophy detectable on brain magnetic resonance imaging (MRI) scans. However, current methods to detect Aβ/tau pathology are either invasive (lumbar puncture) or quite costly and not widely available (positron emission tomography (PET)). And one of the particular neurodegenerative regions is the hippocampus to which the influence of Aβ/tau on has been one of the research projects focuses in the AD pathophysiological progress. In this dissertation, I proposed three novel machine learning and statistical models to examine subtle aspects of the hippocampal morphometry from MRI that are associated with Aβ /tau burden in the brain, measured using PET images. The first model is a novel unsupervised feature reduction model to generate a low-dimensional representation of hippocampal morphometry for each individual subject, which has superior performance in predicting Aβ/tau burden in the brain. The second one is an efficient federated group lasso model to identify the hippocampal subregions where atrophy is strongly associated with abnormal Aβ/Tau. The last one is a federated model for imaging genetics, which can identify genetic and transcriptomic influences on hippocampal morphometry. Finally, I stated the results of these three models that have been published or submitted to peer-reviewed conferences and journals.

Deep neural networks have been shown to be vulnerable to adversarial attacks. Typical attack strategies alter authentic data subtly so as to obtain adversarial samples that resemble the original but otherwise would cause a network's misbehavior such as a high misclassification rate. Various attack approaches have been reported, with some showing state-of-the-art performance in attacking certain networks. In the meanwhile, many defense mechanisms have been proposed in the literature, some of which are quite effective for guarding against typical attacks. Yet, most of these attacks fail when the targeted network modifies its architecture or uses another set of parameters and vice versa. Moreover, the emerging of more advanced deep neural networks, such as generative adversarial networks (GANs), has made the situation more complicated and the game between the attack and defense is continuing. This dissertation aims at exploring the venerability of the deep neural networks by investigating the mechanisms behind the success/failure of the existing attack and defense approaches. Therefore, several deep learning-based approaches have been proposed to study the problem from different perspectives. First, I developed an adversarial attack approach by exploring the unlearned region of a typical deep neural network which is often over-parameterized. Second, I proposed an end-to-end learning framework to analyze the images generated by different GAN models. Third, I developed a defense mechanism that can secure the deep neural network against adversarial attacks with a defense layer consisting of a set of orthogonal kernels. Substantial experiments are conducted to unveil the potential factors that contribute to attack/defense effectiveness. This dissertation also concludes with a discussion of possible future works of achieving a robust deep neural network.

This dissertation constructs a new computational processing framework to robustly and precisely quantify retinotopic maps based on their angle distortion properties. More generally, this framework solves the problem of how to robustly and precisely quantify (angle) distortions of noisy or incomplete (boundary enclosed) 2-dimensional surface to surface mappings. This framework builds upon the Beltrami Coefficient (BC) description of quasiconformal mappings that directly quantifies local mapping (circles to ellipses) distortions between diffeomorphisms of boundary enclosed plane domains homeomorphic to the unit disk. A new map called the Beltrami Coefficient Map (BCM) was constructed to describe distortions in retinotopic maps. The BCM can be used to fully reconstruct the original target surface (retinal visual field) of retinotopic maps. This dissertation also compared retinotopic maps in the visual processing cascade, which is a series of connected retinotopic maps responsible for visual data processing of physical images captured by the eyes. By comparing the BCM results from a large Human Connectome project (HCP) retinotopic dataset (N=181), a new computational quasiconformal mapping description of the transformed retinal image as it passes through the cascade is proposed, which is not present in any current literature. The description applied on HCP data provided direct visible and quantifiable geometric properties of the cascade in a way that has not been observed before. Because retinotopic maps are generated from in vivo noisy functional magnetic resonance imaging (fMRI), quantifying them comes with a certain degree of uncertainty. To quantify the uncertainties in the quantification results, it is necessary to generate statistical models of retinotopic maps from their BCMs and raw fMRI signals. Considering that estimating retinotopic maps from real noisy fMRI time series data using the population receptive field (pRF) model is a time consuming process, a convolutional neural network (CNN) was constructed and trained to predict pRF model parameters from real noisy fMRI data

Machine learning models and in specific, neural networks, are well known for being inscrutable in nature. From image classification tasks and generative techniques for data augmentation, to general purpose natural language models, neural networks are currently the algorithm of preference that is riding the top of the current artificial intelligence (AI) wave, having experienced the greatest boost in popularity above any other machine learning solution. However, due to their inscrutable design based on the optimization of millions of parameters, it is ever so complex to understand how their decision is influenced nor why (and when) they fail. While some works aim at explaining neural network decisions or making systems to be inherently interpretable the great majority of state of the art machine learning works prioritize performance over interpretability effectively becoming black boxes. Hence, there is still uncertainty in the decision boundaries of these already deployed solutions whose predictions should still be analyzed and taken with care. This becomes even more important when these models are used on sensitive scenarios such as medicine, criminal justice, settings with native inherent social biases or where egregious mispredictions can negatively impact the system or human trust down the line. Thus, the aim of this work is to provide a comprehensive analysis on the failure modes of the state of the art neural networks from three domains: large image classifiers and their misclassifications, generative adversarial networks when used for data augmentation and transformer networks applied to structured representations and reasoning about actions and change.

Increased LV wall thickness is frequently encountered in transthoracicechocardiography (TTE). While accurate and early diagnosis is clinically important, given the differences in available therapeutic options and prognosis, an extensive workup is often required for establishing the diagnosis. I propose the first echo-based, automated deep learning model with a fusion architecture to facilitate the evaluation and diagnosis of increased left ventricular (LV) wall thickness. Patients with an established diagnosis for increased LV wall thickness (hypertrophic cardiomyopathy (HCM), cardiac amyloidosis (CA), and hypertensive heart disease (HTN)/others) between 1/2015 to 11/2019 at Mayo Clinic Arizona were identified. The cohort was divided into 80%/10%/10% for training, validation, and testing sets, respectively. Six baseline TTE views were used to optimize a pre-trained InceptionResnetV2 model, each model output was used to train a meta-learner under a fusion architecture. Model performance was assessed by multiclass area under the receiver operating characteristic curve (AUROC). A total of 586 patients were used for the final analysis (194 HCM, 201 CA, and 191 HTN/others). The mean age was 55.0 years, and 57.8% were male. Among the individual view-dependent models, the apical 4 chamber model had the best performance (AUROC: HCM: 0.94, CA: 0.73, and HTN/other: 0.87). The final fusion model outperformed all the view-dependent models (AUROC: CA: 0.90, HCM: 0.93, and HTN/other: 0.92). I successfully established an automatic end-to-end deep learning model framework that accurately differentiates the major etiologies of increased LV wall thickness, including HCM and CA from the background of HTN/other diagnoses.

The omnipresent data, growing number of network devices, and evolving attack techniques have been challenging organizations’ security defenses over the past decade. With humongous volumes of logs generated by those network devices, looking for patterns of malicious activities and identifying them in time is growing beyond the capabilities of their defense systems. Deep Learning, a subset of Machine Learning (ML) and Artificial Intelligence (AI), fills in this gapwith its ability to learn from huge amounts of data, and improve its performance as the data it learns from increases. In this dissertation, I bring forward security issues pertaining to two top threats that most organizations fear, Advanced Persistent Threat (APT), and Distributed Denial of Service (DDoS), along with deep learning models built towards addressing those security issues. First, I present a deep learning model, APT Detection, capable of detecting anomalous activities in a system. Evaluation of this model demonstrates how it can contribute to early detection of an APT attack with an Area Under the Curve (AUC) of up to 91% on a Receiver Operating Characteristic (ROC) curve. Second, I present DAPT2020, a first of its kind dataset capturing an APT attack exploiting web and system vulnerabilities in an emulated organization’s production network. Evaluation of the dataset using well known machine learning models demonstrates the need for better deep learning models to detect APT attacks. I then present DAPT2021, a semi-synthetic dataset capturing an APT attackexploiting human vulnerabilities, alongside 2 less skilled attacks. By emulating the normal behavior of the employees in a set target organization, DAPT2021 has been created to enable researchers study the causations and correlations among the captured data, a much-needed information to detect an underlying threat early. Finally, I present a distributed defense framework, SmartDefense, that can detect and mitigate over 90% of DDoS traffic at the source and over 97.5% of the remaining DDoS traffic at the Internet Service Provider’s (ISP’s) edge network. Evaluation of this work shows how by using attributes sent by customer edge network, SmartDefense can further help ISPs prevent up to 51.95% of the DDoS traffic from going to the destination.

How to teach a machine to understand natural language? This question is a long-standing challenge in Artificial Intelligence. Several tasks are designed to measure the progress of this challenge. Question Answering is one such task that evaluates a machine's ability to understand natural language, where it reads a passage of text or an image and answers comprehension questions. In recent years, the development of transformer-based language models and large-scale human-annotated datasets has led to remarkable progress in the field of question answering. However, several disadvantages of fully supervised question answering systems have been observed. Such as generalizing to unseen out-of-distribution domains, linguistic style differences in questions, and adversarial samples. This thesis proposes implicitly supervised question answering systems trained using knowledge acquisition from external knowledge sources and new learning methods that provide inductive biases to learn question answering. In particular, the following research projects are discussed: (1) Knowledge Acquisition methods: these include semantic and abductive information retrieval for seeking missing knowledge, a method to represent unstructured text corpora as a knowledge graph, and constructing a knowledge base for implicit commonsense reasoning. (2) Learning methods: these include Knowledge Triplet Learning, a method over knowledge graphs; Test-Time Learning, a method to generalize to an unseen out-of-distribution context; WeaQA, a method to learn visual question answering using image captions without strong supervision; WeaSel, weakly supervised method for relative spatial reasoning; and a new paradigm for unsupervised natural language inference. These methods potentially provide a new research direction to overcome the pitfalls of direct supervision.