Despite the impressive advancements in people detection and tracking, safety is still a key barrier to the deployment of autonomous vehicles in urban environments [1]. For example, in non-autonomous technology, there is an implicit communication between the people crossing the street and the driver to make sure they have communicated their intent to the driver. Therefore, it is crucial for the autonomous car to infer the future intent of the pedestrian quickly. We believe that human body orientation with respect to the camera can help the intelligent unit of the car to anticipate the future movement of the pedestrians. To further improve the safety of pedestrians, it is important to recognize whether they are distracted, carrying a baby, or pushing a shopping cart. Therefore, estimating the fine- grained 3D pose, i.e. (x,y,z)-coordinates of the body joints provides additional information for decision-making units of driverless cars. In this dissertation, we have proposed a deep learning-based solution to classify the categorized body orientation in still images. We have also proposed an efficient framework based on our body orientation classification scheme to estimate human 3D pose in monocular RGB images. Furthermore, we have utilized the dynamics of human motion to infer the body orientation in image sequences. To achieve this, we employ a recurrent neural network model to estimate continuous body orientation from the trajectories of body joints in the image plane. The proposed body orientation and 3D pose estimation framework are tested on the largest 3D pose estimation benchmark, Human3.6m (both in still images and video), and we have proved the efficacy of our approach by benchmarking it against the state-of-the-art approaches. Another critical feature of self-driving car is to avoid an obstacle. In the current prototypes the car either stops or changes its lane even if it causes other traffic disruptions. However, there are situations when it is preferable to collide with the object, for example a foam box, rather than take an action that could result in a much more serious accident than collision with the object. In this dissertation, for the first time, we have presented a novel method to discriminate between physical properties of these types of objects such as bounciness, elasticity, etc. based on their motion characteristics . The proposed algorithm is tested on synthetic data, and, as a proof of concept, its effectiveness on a limited set of real-world data is demonstrated.

Feature engineering is a process that augments the feature vector of a machine learning model with calculated values that are designed to enhance the accuracy of a model’s predictions. Research has shown that the accuracy of models such as deep neural networks, support vector machines, and tree/forest-based algorithms sometimes benefit from feature engineering. Expressions that combine one or more of the original features usually create these engineered features. The choice of the exact structure of an engineered feature is dependent on the type of machine learning model in use. Previous research demonstrated that various model families benefit from different types of engineered feature. Random forests, gradient-boosting machines, or other tree-based models might not see the same accuracy gain that an engineered feature allowed neural networks, generalized linear models, or other dot-product based models to achieve on the same data set. This dissertation presents a genetic programming-based algorithm that automatically engineers features that increase the accuracy of deep neural networks for some data sets. For a genetic programming algorithm to be effective, it must prioritize the search space and efficiently evaluate what it finds. This dissertation algorithm faced a potential search space composed of all possible mathematical combinations of the original feature vector. Five experiments were designed to guide the search process to efficiently evolve good engineered features. The result of this dissertation is an automated feature engineering (AFE) algorithm that is computationally efficient, even though a neural network is used to evaluate each candidate feature. This approach gave the algorithm a greater opportunity to specifically target deep neural networks in its search for engineered features that improve accuracy. Finally, a sixth experiment empirically demonstrated the degree to which this algorithm improved the accuracy of neural networks on data sets augmented by the algorithm’s engineered features.

A security operation center (SOC) is a cybersecurity clearinghouse responsible for monitoring, collecting and analyzing security events from organizations’ IT infrastructure and security controls. Despite their popularity, SOCs are facing increasing challenges and pressure due to the growing volume, velocity and variety of the IT infrastructure and security data observed on a daily basis. Due to the mixed performance of current technological solutions, e.g. intrusion detection system (IDS) and security information and event management (SIEM), there is an over-reliance on manual analysis of the events by human security analysts. This creates huge backlogs and slows down considerably the resolution of critical security events. Obvious solutions include increasing the accuracy and efficiency of crucial aspects of the SOC automation workflow, such as the event classification and prioritization. In the current thesis, we present a new approach for SOC event classification and prioritization by identifying a set of new machine learning features using graph visualization and graph metrics. Using a real-world SOC dataset and by applying different machine learning classification techniques, we demonstrate empirically the benefit of using the graph-based features in terms of improved classification accuracy. Three different classification techniques are explored, namely, logistic regression, XGBoost and deep neural network (DNN). The experimental evaluation shows for the DNN, the best performing classifier, area under curve (AUC) values of 91% for the baseline feature set and 99% for the augmented feature set that includes the graph-based features, which is a net improvement of 8% in classification performance. / Graduate

<p>Every optimization problem shares the common objective of finding a minima/maxima, but its application spans over a wide variety of fields ranging from solving NP-hard problems to training a neural network. This thesis addresses two crucial aspects of the above-mentioned fields. The first project is concerned with designing a hardware-system for efficiently solving Traveling Salesman Problem (TSP). It involves encoding  the solution to the ground state of an Ising Hamiltonian and finding the minima of the energy landscape. To that end, we i) designed a stochastic nanomagnet-based device as a building block for the system, ii) developed a unique approach to encode any TSP into an array of these blocks, and finally, iii) established the operating principle to make the system converge to an optimal solution. We used this method to solve TSPs having more than 600 nodes.</p> <p>  </p> <p>The next parts of the thesis deal with another genre of optimization problems involving deep neural networks (DNN) in image-classification tasks. DNNs are trained by finding the minima of a loss landscape aimed at mapping input images to a set of discrete labels. Adversarial attacks tend to disrupt this mapping by corrupting the inputs with subtle perturbations, imperceptible to human eyes. Although it is imperative to deploy some external defense mechanisms to guard against these attacks, the defense procedure can be aided by some intrinsic robust properties of the network. In the quest for an inherently resilient neural network, we explored the robustness of biologically-inspired Spiking Neural Networks (SNN) in the second part of the thesis. We demonstrated that accuracy degradation is less severe in SNNs than in their non-spiking counterparts. We attribute this robustness to two fundamental characteristics of SNNs: (i) input discretization  and (ii) leak rate in Leaky-Integrate-Fire neurons and analyze their effects.</p> <p><br></p> <p>As mentioned beforehand, this intrinsic robustness is merely an aiding tool to external defense mechanisms. Adversarial training has been established as the stat-of-the-art defense to provide significant robustness against existing attack techniques. This method redefines the boundary of the neural network by augmenting the training dataset with adversarial samples. In the process of achieving robustness, we are faced with a trade-off: a decrease in the prediction accuracy of clean or unperturbed data. The goal of the last section of my thesis is to understand this setback by using Gradient Projection-based sequential learning as an analysis tool. We systematically analyze the interplay between clean training and adversarial training on parameter subspace. In this technique, adversarial training follows clean training task where the parameter update is performed in the orthogonal direction of the previous task (clean training). It is possible to track down the principal component directions responsible for adversarial training by restricting clean and adversarial parameter update to two orthogonal subspaces. By varying the partition of subspace, we showed that the low-variance principal components are not capable of learning adversarial data, rather it is necessary to perform parameter update in a common subspace consisting of higher variance principal components to obtain significant adversarial accuracy. However, disturbing these higher variance components causes the decrease in standard clean accuracy, hence the accuracy-robustness trade-off. Further, we showed that this trade-off is worsened</p> <p>when the network capacity is smaller due to under-parameterization effect.</p>

UAV (Unmanned Airborne Vehicle) is a source of entertainment and a pleasurable experience, attracting many young people to pursue it as a hobby. With the potential increase in the number of UAVs, the risk of using them for malicious purposes also increases. In addition, birds and UAVs have very similar maneuvers during flights. These UAVs can also carry a significant payload, which can have unintended consequences. Therefore, detecting UAVs near red-zone areas is an important problem. In addition, small UAVs can record video from large distances without being spotted by the naked eye. An appropriate network of sensors may be needed to foresee the arrival of such entities from a safe distance before they pose any danger to the surrounding areas. Despite the growing interest in UAV detection, limited research has been conducted in this area due to a lack of available data for model training. This thesis proposes a novel approach to address this challenge by leveraging experimental data collected in real-time using high-sensitivity sensors instead of relying solely on simulations. This approach allows for improved model accuracy and a better representation of the complex and dynamic environments in which UAVs operate, which are difficult to simulate accurately. The thesis further explores the application of machine learning and sensor fusion algorithms to detect UAVs and distinguish them from other objects, such as birds, in real-time. Specifically, the thesis utilizes YOLOv3 with deep sort and sensor fusion algorithms to achieve accurate UAV detection. In this study, we employed YOLOv3, a deep learning model known for its high efficiency and complexity, to facilitate real-time drone versus bird detection. To further enhance the reliability of the system, we incorporated sensor fusion, leading to a more stable and accurate real-time system, and mitigating the incidence of false detections. Our study indicates that the YOLOv3 model outperformed the state-of-the-art models in terms of both speed and robustness, achieving a high level of confidence with a score above 95%. Moreover, the YOLOv3 model demonstrated a promising capability in real-time drone versus bird detection, which suggests its potential for practical applications

Yes / Burns are one of the obnoxious injuries subjecting thousands to loss of life and physical defacement each year. Both high income and Third World countries face major evaluation challenges including but not limited to inadequate workforce, poor diagnostic facilities, inefficient diagnosis and high operational cost. As such, there is need to develop an automatic machine learning algorithm to noninvasively identify skin burns. This will operate with little or no human intervention, thereby acting as an affordable substitute to human expertise. We leverage the weights of pretrained deep neural networks for image description and, subsequently, the extracted image features are fed into the support vector machine for classification. To the best of our knowledge, this is the first study that investigates black African skins. Interestingly, the proposed algorithm achieves state-of-the-art classification accuracy on both Caucasian and African datasets.

Dialectal Arabic speech research suffers from the lack of labelled resources and standardised orthography. There are three main challenges in dialectal Arabic speech recognition: (i) finding labelled dialectal Arabic speech data, (ii) training robust dialectal speech recognition models from limited labelled data and (iii) evaluating speech recognition for dialects with no orthographic rules. This thesis is concerned with the following three contributions: Arabic Dialect Identification: We are mainly dealing with Arabic speech without prior knowledge of the spoken dialect. Arabic dialects could be sufficiently diverse to the extent that one can argue that they are different languages rather than dialects of the same language. We have two contributions: First, we use crowdsourcing to annotate a multi-dialectal speech corpus collected from Al Jazeera TV channel. We obtained utterance level dialect labels for 57 hours of high-quality consisting of four major varieties of dialectal Arabic (DA), comprised of Egyptian, Levantine, Gulf or Arabic peninsula, North African or Moroccan from almost 1,000 hours. Second, we build an Arabic dialect identification (ADI) system. We explored two main groups of features, namely acoustic features and linguistic features. For the linguistic features, we look at a wide range of features, addressing words, characters and phonemes. With respect to acoustic features, we look at raw features such as mel-frequency cepstral coefficients combined with shifted delta cepstra (MFCC-SDC), bottleneck features and the i-vector as a latent variable. We studied both generative and discriminative classifiers, in addition to deep learning approaches, namely deep neural network (DNN) and convolutional neural network (CNN). In our work, we propose Arabic as a five class dialect challenge comprising of the previously mentioned four dialects as well as modern standard Arabic. Arabic Speech Recognition: We introduce our effort in building Arabic automatic speech recognition (ASR) and we create an open research community to advance it. This section has two main goals: First, creating a framework for Arabic ASR that is publicly available for research. We address our effort in building two multi-genre broadcast (MGB) challenges. MGB-2 focuses on broadcast news using more than 1,200 hours of speech and 130M words of text collected from the broadcast domain. MGB-3, however, focuses on dialectal multi-genre data with limited non-orthographic speech collected from YouTube, with special attention paid to transfer learning. Second, building a robust Arabic ASR system and reporting a competitive word error rate (WER) to use it as a potential benchmark to advance the state of the art in Arabic ASR. Our overall system is a combination of five acoustic models (AM): unidirectional long short term memory (LSTM), bidirectional LSTM (BLSTM), time delay neural network (TDNN), TDNN layers along with LSTM layers (TDNN-LSTM) and finally TDNN layers followed by BLSTM layers (TDNN-BLSTM). The AM is trained using purely sequence trained neural networks lattice-free maximum mutual information (LFMMI). The generated lattices are rescored using a four-gram language model (LM) and a recurrent neural network with maximum entropy (RNNME) LM. Our official WER is 13%, which has the lowest WER reported on this task. Evaluation: The third part of the thesis addresses our effort in evaluating dialectal speech with no orthographic rules. Our methods learn from multiple transcribers and align the speech hypothesis to overcome the non-orthographic aspects. Our multi-reference WER (MR-WER) approach is similar to the BLEU score used in machine translation (MT). We have also automated this process by learning different spelling variants from Twitter data. We mine automatically from a huge collection of tweets in an unsupervised fashion to build more than 11M n-to-m lexical pairs, and we propose a new evaluation metric: dialectal WER (WERd). Finally, we tried to estimate the word error rate (e-WER) with no reference transcription using decoding and language features. We show that our word error rate estimation is robust for many scenarios with and without the decoding features.

Recent discoveries uncovered flaws in machine learning algorithms such as deep neural networks. Deep neural networks seem vulnerable to small amounts of non-random noise, created by exploiting the input to output mapping of the network. Applying this noise to an input image drastically decreases classication performance. Such image is referred to as an adversarial example. The purpose of this thesis is to examine how known regularization/robustness methods perform on adversarial examples. The robustness methods: dropout, low-pass filtering, denoising autoencoder, adversarial training and committees have been implemented, combined and tested. For the well-known benchmark, the MNIST (Mixed National Institute of Standards and Technology) dataset, the best combination of robustness methods has been found. Emerged from the results of the experiments, ensemble of models trained on adversarial examples is considered to be the best approach for MNIST. Harmfulness of the adversarial noise and some robustness experiments are demonstrated on CIFAR10 (The Canadian Institute for Advanced Research) dataset as well. Apart from robustness tests, the thesis describes experiments with human classification performance on noisy images and the comparison with performance of deep neural network.

Across domains, the scale of data and complexity of models have both been increasing greatly in the recent years. For many models of interest, tractable learning and inference without access to expensive computational resources have become challenging. In this thesis, we approach efficient learning and inference through the leverage of sparse structures inherent in the learning objective, which allows us to develop algorithms sublinear in the size of parameters without compromising the accuracy of models. In particular, we address the following three questions for each problem of interest: (a) how to formulate model estimation as an optimization problem with tractable sparse structure, (b) how to efficiently, i.e. in sublinear time, search, maintain, and utilize the sparse structures during training and inference, (c) how to guarantee fast convergence of our optimization algorithm despite its greedy nature? By answering these questions, we develop state-of-the-art algorithms in varied domains. Specifically, in the extreme classification domain, we utilizes primal and dual sparse structures to develop greedy algorithms of complexity sublinear in the number of classes, which obtain state-of-the-art accuracies on several benchmark data sets with one or two orders of magnitude speedup over existing algorithms. We also apply the primal-dual-sparse theory to develop a state-of-the-art trimming algorithm for Deep Neural Networks, which sparsifies neuron connections of a DNN with a task-dependent theoretical guarantee, which results in models of smaller storage cost and faster inference speed. When it comes to structured prediction problems (i.e. graphical models) with inter-dependent outputs, we propose decomposition methods that exploit sparse messages to decompose a structured learning problem of large output domains into factorwise learning modules amenable to sublineartime optimization methods, leading to practically much faster alternatives to existing learning algorithms. The decomposition technique is especially effective when combined with search data structures, such as those for Maximum Inner-Product Search (MIPS), to improve the learning efficiency jointly. Last but not the least, we design novel convex estimators for a latent-variable model by reparameterizing it as a solution of sparse support in an exponentially high-dimensional space, and approximate it with a greedy algorithm, which yields the first polynomial-time approximation method for the Latent-Feature Models and Generalized Mixed Regression without restrictive data assumptions.

Autonomous relative terrain navigation is a problem at the forefront of many space missions involving close proximity operations to any target body. With no definitive answer, there are many techniques to help cope with this issue using both passive and active sensors, but almost all require high fidelity models of the associated dynamics in the environment. Convolutional Neural Networks (CNNs) trained with images rendered from a digital terrain map (DTM) of the body’s surface can provide a way to side-step the issue of unknown or complex dynamics while still providing reliable autonomous navigation. This is achieved by directly mapping an image to a relative position to the target body. The portability of trained CNNs allows “offline” training that can yield a matured network capable of being loaded onto a spacecraft for real-time position acquisition. In this thesis the lunar surface is used as the proving ground for this optical navigation technique, but the methods used are not unique to the Moon, and are applicable in general.