<div> <div> <div> <p>Proteins are drivers of almost all biological processes in the cell. The functions of a protein are dependent on their three-dimensional structure and elucidating the structure and function of proteins is key to understanding how a biological system operates. In this research, we developed computational methods using machine learning techniques to predicts the structure and function of proteins. Protein 3D structure prediction has advanced significantly in recent years, largely due to deep learning approaches that predict inter-residue contacts and, more recently, distances using multiple sequence alignments (MSAs). The performance of these models depends on the number of similar protein sequences to the query protein, wherein some cases similar sequences are few but dissimilar sequences with local similarities are more and can be helpful. We have developed a novel deep learning-based approach AttentiveDist which further improves over the previous state of art. We added an attention mechanism where dis-similar sequences are also used (increasing number of sequences) and the model itself determines which information from such sequences it should attend to. We showed that the improvement of distance predictions was successfully transferred to achieve better protein tertiary structure modeling. We also show that structure prediction from a predicted distance map can be further enhanced by using predicted inter-residue sidechain center distances and main-chain hydrogen-bonds. Protein function prediction is another avenue we explored where we want to predict the function that a protein will perform. The crux of the approach is to predict the function of protein based on the function of similar sequences. Here, we developed a method where we use dissimilar sequences to extract additional information and improve performance over the previous approaches. We used phylogenetic analysis to determine if a dissimilar sequence can be close to the query sequence and thus can provide functional information. Our method was ranked highly in worldwide protein function prediction competition CAFA3 (2016-2019). Further, we expanded the method with a neural network to predict protein toxicity that can be used as a safety check for human-designed protein sequences.</p></div></div></div>

<p>Digital image processing techniques have many significant applications in industry. In this thesis, we focus on three problems in digital image processing. These three problems involve halftone images, information encoding and decoding, image alignment, and deep learning.</p><p>Specifically, the first problem is based on data-bearing halftone images, which are an aesthetically pleasing alternative to barcodes. We address the issues generated in the camera captured image alignment process. We perform some theoretical analysis and validate it by simulation. We also provide an optimal solution to the problem.</p><p>The second problem is about the alignment technique on a 3D surface. We develop a pipeline of surfaces coding to solve the alignment issues on 3D surfaces, which includes oblique surfaces and cylindrical surfaces.</p><p>The third problem is related to image retrieval. We propose a deep learning based solution to the fashion image retrieval task. Fashion image retrieval is significant to improve the customers’ experience in online shopping. A fast, accurate shopping item information retrieval system based on the customers’ uploaded image has been built by us. A novel solution is provided, and it achieves state-of-art accuracy in shopping items’ information retrieval.</p>

<p>The rapid expansion of intra-vehicle networks has increased the number of threats to such networks. Most modern vehicles implement various physical and data-link layer technologies. Vehicles are becoming increasingly autonomous and connected. Controller Area Network (CAN) is a serial bus system that is used to connect sensors and controllers (Electronic Control Units – ECUs) within a vehicle. ECUs vary widely in processing power, storage, memory, and connectivity. The goal of this research is to design, implement, and test an efficient and effective intrusion detection system for intra-vehicle CANs. Such a system must be capable of detecting intrusions in almost real-time with minimal resources. The research proposes a specific type of recursive neural network called Long Short-Term Memory (LSTM) to detect anomalies. It also proposes a decision engine that will use LSTM-classified anomalies to detect intrusions by using multiple contextual parameters. We have conducted multiple experiments on the optimal choice of various LSTM hyperparameters. We have tested our classification algorithm and our decision engine using data from real automobiles. We will present the results of our experiments and analyze our findings. After detailed evaluation of our intrusion detection system, we believe that we have designed a vehicle security solution that meets all the outlined requirements and goals.</p>

<p>  </p> <p>Smart manufacturing is a broad concept of manufacturing technology that employs the computer aided systems, digital information technology, artificially intelligent algorithms, etc., to realize high-level automation of the production. The rise of the smart manufacturing concept, which has also been treated as the fourth industrial revolution, has been increasingly advocated by the policy makers and investigated by the worldwide researchers. Though machining is one of the key processes in the manufacturing industry, there are only a few researches focusing on automatically scheduling and improving the machining process. The design of the machining parameters and tool path planning still requires engineers with significant knowledge and experience in manufacturing fields to juggle between product quality, machine tool maintenance, and production cost. This design process also requires high level of human intelligence to consider the type of material, machine tool setups, workpiece geometry, and cutting tool property to provide an optimal manufacturing process. The overall machining related processes cannot satisfy the requirement of the ultimate goal of the smart manufacturing – to fully automate the machining process without human’s involvements.</p> <p><br></p> <p>In order to solve this problem, we aim to employ advanced machine learning technologies to enable the machine tool to automatically build up the cutting physics and generate the optimized toolpath. The final optimized result can be conducted automatically and shows a near human level optimization design ability. The generated toolpath beats the result from other commercial software. The overall framework can be fully automated when the machine learning technology is mature. </p>

<p>The design and discovery of novel materials are difficult not only due to expensive and time- consuming calculation and measurements of their properties, but also thanks to the infinite search spaces. With the increasingly abundant data from experiments and simulations, learning from data has the potential of bypassing complex physics-based simulations and experiments and providing fast approximations of the solution. Deep learning models are helpful in the design process that requires prohibitively expensive iterative computations. In addition, as efficient and accurate sur- rogate models, trained deep networks can incorporate techniques, such as sensitivity analysis and active learning, to provide guidance in searching promising candidates. Moreover, deep learning models need to account for the material structural information, such as molecule and atom align- ments, chemical bonds, and grain-level interactions, as it plays an important role in determining the macroscopic properties. In this thesis, we start with developing two standard deep learning model- based materials design frameworks for lithium-ion batteries and thermoelectric materials, and we then investigate the feasibility of standard deep learning models on data with graph-structured in- formation and identify the challenges. Finally, we propose a deep graph operator network that effectively capture the spatial dependency encoded in the graph structure to solve networked dy- namical systems.</p> <p><br></p> <p>In the first half of the thesis, we propose a hybrid convolutional neural network to infer lithium- ion battery microstructure properties, Bruggeman’s exponent and shape factor, given its voltage vs. capacity curves. The trained model accurately predicts the microstructural properties on both experimental and simulation data, and it can readily accelerate the processing-properties- performance and degradation characteristics of the existing and emerging chemistries of lithium- ion batteries. Also, we develop a AI-guided framework to discover and design thermoelectric materials, where we train classifiers based on the materials chemical and structural information embeddings and combine with variance-based sensitivity analysis to suggest candidates and con- duct fast screening.</p> <p><br></p> <p>In the second half of the thesis, we build a data-centric framework with a recurrent neural network-based classifier to achieve traffic incident detection on highway networks. We incorporate weak supervised learning and design labeling functions to create large amount of training data with probabilistic labels. The trained deep ensemble accurately detects incidents with predictive uncertainty. To capture the structural information in the network, we then propose a deep graph operator network that maps the input graph state function to the output graph state function. The proposed model enables resolution-independence and zero-shot transfer, where we do not require a set of fixed sensors to encode the graph trajectory and can use the trained model directly on larger graphs with high accuracy. We utilize the proposed model to solve power grid transient stability prediction and traffic forecasting problems.</p>

<p>Temporal point processes (TPP) are mathematical approaches for modeling asynchronous event sequences by considering the temporal dependency of each event on past events and its instantaneous rate. Temporal point processes can model various problems, from earthquake aftershocks, trade orders, gang violence, and reported crime patterns, to network analysis, infectious disease transmissions, and virus spread forecasting. In each of these cases, the entity's behavior with the corresponding information is noted over time as an asynchronous event sequence, and the analysis is done using temporal point processes, which provides a means to define the generative mechanism of the sequence of events and ultimately predict events and investigate causality.</p> <p><br></p> <p>Among point processes, Hawkes process as a stochastic point process is able to model a wide range of contagious and self-exciting patterns. One of Hawkes process's well-known applications is predicting the evolution of viral processes on networks, which is an important problem in biology, the social sciences, and the study of the Internet. In existing works, mean-field analysis based upon degree distribution is used to predict viral spreading across networks of different types. However, it has been shown that degree distribution alone fails to predict the behavior of viruses on some real-world networks. Recent attempts have been made to use assortativity to address this shortcoming. This thesis illustrates how the evolution of such a viral process is sensitive to the underlying network's structure. </p> <p><br></p> <p>In Chapter 3, we show that adding assortativity does not fully explain the variance in the spread of viruses for a number of real-world networks. We propose using the graphlet frequency distribution combined with assortativity to explain variations in the evolution of viral processes across networks with identical degree distribution. Using a data-driven approach, by coupling predictive modeling with viral process simulation on real-world networks, we show that simple regression models based on graphlet frequency distribution can explain over 95\% of the variance in virality on networks with the same degree distribution but different network topologies. Our results highlight the importance of graphlets and identify a small collection of graphlets that may have the most significant influence over the viral processes on a network.</p> <p><br></p> <p>Due to the flexibility and expressiveness of deep learning techniques, several neural network-based approaches have recently shown promise for modeling point process intensities. However, there is a lack of research on the possible adversarial attacks and the robustness of such models regarding adversarial attacks and natural shocks to systems. Furthermore, while neural point processes may outperform simpler parametric models on in-sample tests, how these models perform when encountering adversarial examples or sharp non-stationary trends remains unknown. </p> <p><br></p> <p>In Chapter 4, we propose several white-box and black-box adversarial attacks against deep temporal point processes. Additionally, we investigate the transferability of white-box adversarial attacks against point processes modeled by deep neural networks, which are considered a more elevated risk. Extensive experiments confirm that neural point processes are vulnerable to adversarial attacks. Such a vulnerability is illustrated both in terms of predictive metrics and the effect of attacks on the underlying point process's parameters. Expressly, adversarial attacks successfully transform the temporal Hawkes process regime from sub-critical to into a super-critical and manipulate the modeled parameters that is considered a risk against parametric modeling approaches. Additionally, we evaluate the vulnerability and performance of these models in the presence of non-stationary abrupt changes, using the crimes and Covid-19 pandemic dataset as an example.</p> <p><br></p> <p> Considering the security vulnerability of deep-learning models, including deep temporal point processes, to adversarial attacks, it is essential to ensure the robustness of the deployed algorithms that is despite the success of deep learning techniques in modeling temporal point processes.</p> <p> </p> <p>In Chapter 5, we study the robustness of deep temporal point processes against several proposed adversarial attacks from the adversarial defense viewpoint. Specifically, we investigate the effectiveness of adversarial training using universal adversarial samples in improving the robustness of the deep point processes. Additionally, we propose a general point process domain-adopted (GPDA) regularization, which is strictly applicable to temporal point processes, to reduce the effect of adversarial attacks and acquire an empirically robust model. In this approach, unlike other computationally expensive approaches, there is no need for additional back-propagation in the training step, and no further network is required. Ultimately, we propose an adversarial detection framework that has been trained in the Generative Adversarial Network (GAN) manner and solely on clean training data. </p> <p><br></p> <p>Finally, in Chapter 6, we discuss implications of the research and future research directions.</p>

<div>Artificial Intelligence (AI) plays an increasingly pivotal role in drug discovery. In particular, artificial neural networks such as deep neural networks drive this area of research. The research presented in this thesis is considered a synergistic combination of physicochemical models of protein-ligand interactions such as molecular dynamics simulation, novel machine learning concepts and the use of big data for solving fundamental problems in Structure-Based Drug Design (SBDD). This area of research involves the use of three-dimensional (3D) structural data of biomolecules to assist lead discovery and optimization in a time- and cost-efficient manner. </div><div>The main focus of the thesis research is the development of models, algorithms and methods to facilitate binding-mode elucidation, affinity prediction for congeneric series of molecules and flexible docking. </div><div><br></div><div>For pose-prediction, we developed a Convolutional Neural Network model incorporating hydration information, named DeepWatsite, which displays accurate binding-mode prediction and the capability to highlight different roles of water molecules in protein-ligand binding. In order to train the neural network model, we created a comprehensive database for hydration information of thousands of protein systems. This was made possible through the development of an efficient GPU-accelerated version of Watsite, a program for generating hydration profiles of protein systems through molecular dynamics simulations.\newline</div><div>\indent For accurate affinity prediction for congeneric series of compounds, we developed a new methodological platform for mixed-solvent simulation based on the lambda-dynamics concept. Additionally, we developed a deep-learning model that combines molecular dynamics simulations and a distance-aware graph attention algorithm. Validation studies using this method revealed that its accuracy is competitive to resource-intensive free energy perturbation (FEP) calculations. To train the model, we generated a synthetic database of congeneric series of compounds extracted from the highest-quality medicinal chemistry articles. Molecular-dynamics simulations were used to simulate all the generated systems as method for data augmentation.\newline</div><div>\indent For flexible docking, we developed a machine-learning assisted docking strategy that relies on protein-ligand distance matrix predictions. This technique is built upon Weisfeiler-Lehman neural network concept with an attention mechanism. Comprehensive validation on docking and cross-docking datasets demonstrated the potential of this method to become a docking concept with higher accuracy and efficiency than existing state-of-the-art flexible docking techniques. </div><div><br></div><div>In summary, the thesis proved the general applicability of deep-learning to various tasks in SBDD. Furthermore, it demonstrates that treating biomolecules as dynamic entities can improve the quality of computational methods in structure-based drug design.</div>

<p>Terminal airspace around an airport is the biggest bottleneck for commercial operations in the National Airspace System (NAS). In order to prognosticate the safety status of the terminal airspace, effective prediction of the airspace evolution is necessary. While there are fixed procedural structures for managing operations at an airport, the confluence of a large number of aircraft and the complex interactions between the pilots and air traffic controllers make it challenging to predict its evolution. Modeling the high-dimensional spatio-temporal interactions in the airspace given different environmental and infrastructural constraints is necessary for effective predictions of future aircraft trajectories that characterize the airspace state at any given moment. A novel deep learning architecture using Graph Neural Networks is proposed to predict trajectories of aircraft 10 minutes into the future and estimate prog?nostic metrics for the airspace. The uncertainty in the future is quantified by predicting distributions of future trajectories instead of point estimates. The framework’s viability for trajectory prediction and prognosis is demonstrated with terminal airspace data from Dallas Fort Worth International Airport (DFW). </p>

<div>The growth in online goods delivery is causing a dramatic surge in urban vehicle traffic from last-mile deliveries. On the other hand, ride-sharing has been on the rise with the success of ride-sharing platforms and increased research on using autonomous vehicle technologies for routing and matching. The future of urban mobility for passengers and goods relies on leveraging new methods that minimize operational costs and environmental footprints of transportation systems. </div><div><br></div><div>This paper considers combining passenger transportation with goods delivery to improve vehicle-based transportation. Even though the problem has been studied with model-based approaches where the dynamic model of the transportation system environment is defined, model-free approaches where the dynamics of the environment are learned by interaction have been demonstrated to be adaptable to new or erratic environment dynamics. </div><div><br></div><div>FlexPool is a distributed model-free deep reinforcement learning algorithm that jointly serves passengers \& goods workloads by learning optimal dispatch policies from its interaction with the environment. The model-free algorithm (as opposed to a model-based one) is an algorithm which does not use the transition probability distribution (and the reward function) associated with the Markov decision process (MDP).</div><div> The proposed algorithm pools passengers for a ride-sharing service and delivers goods using a multi-hop routing method. These flexibilities decrease the fleet's operational cost and environmental footprint while maintaining service levels for passengers and goods. The dispatching algorithm based on deep reinforcement learning is integrated with an efficient matching algorithm for passengers and goods. Through simulations on a realistic urban mobility platform, we demonstrate that FlexPool outperforms other model-free settings in serving the demands from passengers \& goods. FlexPool achieves 30\% higher fleet utilization and 35\% higher fuel efficiency in comparison to (i) model-free approaches where vehicles transport a combination of passengers \& goods without the use of multi-hop transit, and (ii) model-free approaches where vehicles exclusively transport either passengers or goods. </div>

<div>This dissertation addresses the problem of how to design a convolutional neural network (CNN) for giving semantic labels to the points on the ground given the satellite image coverage over the area and, for the ground truth, given the noisy labels in OpenStreetMap (OSM). This problem is made challenging by the fact that -- (1) Most of the images are likely to have been recorded from off-nadir viewpoints for the area of interest on the ground; (2) The user-supplied labels in OSM are frequently inaccurate and, not uncommonly, entirely missing; and (3) The size of the area covered on the ground must be large enough to possess any engineering utility. As this dissertation demonstrates, solving this problem requires that we first construct a DSM (Digital Surface Model) from a stereo fusion of the available images, and subsequently use the DSM to map the individual pixels in the satellite images to points on the ground. That creates an association between the pixels in the images and the noisy labels in OSM. The CNN-based solution we present yields a 4-8% improvement in the per-class segmentation IoU (Intersection over Union) scores compared to the traditional approaches that use the views independently of one another. The system we present is end-to-end automated, which facilitates comparing the classifiers trained directly on true orthophotos vis-`a-vis first training them on the off-nadir images and subsequently translating the predicted labels to geographical coordinates. This work also presents, for arguably the first time, an in-depth discussion of large-area image alignment and DSM construction using tens of true multi-date and multi-view WorldView-3 satellite images on a distributed OpenStack cloud computing platform.</div>