Search

We introduce the problem of model-extraction attacks in cyber-physical systems in which an attacker attempts to estimate (or extract) the feedback controller of the system. Extracting (or estimating) the controller provides an unmatched edge to attackers since it allows them to predict the future control actions of the system and plan their attack accordingly. Hence, it is important to understand the ability of the attackers to perform such an attack. In this paper, we focus on the setting when a Deep Neural Network (DNN) controller is trained using Reinforcement Learning (RL) algorithms and is used to control a stochastic system. We play the role of the attacker that aims to estimate such an unknown DNN controller, and we propose a two-phase algorithm. In the first phase, also called the offline phase, the attacker uses side-channel information about the RL-reward function and the system dynamics to identify a set of candidate estimates of the unknown DNN. In the second phase, also called the online phase, the attacker observes the behavior of the unknown DNN and uses these observations to shortlist the set of final policy estimates. We provide theoretical analysis of the error between the unknown DNN and the estimated one. We also provide numerical results showing the effectiveness of the proposed algorithm.

This dissertation explores how computer science became a discipline in the United States and how this process of discipline formation and institutionalization shaped the content and culture of computer science. Like other fields of knowledge, computer science in the United States emerged out of established disciplines; however, unlike many other scientific or engineering fields, computer science developed simultaneously out of multiple other disciplines and multiple different interdisciplinary arrangements. This dissertation explores how each of these disciplines in specific university contexts contributed to the construction of the character of computer science and the definition of the identity of the computer scientist and how different interdisciplinary relationships provided opportunities for the development of distinct visions of computer science. Drawing on archival research from university and department records as well as the papers of university administrators, computer scientists, mathematicians, electrical engineers, and others, this work reveals how the processes of discipline formation shaped everything from the intellectual foundation and internal culture of computer science to the modern American university’s institutional structure and approach to interdisciplinary knowledge-making.

Real-time systems are the canonical class of applications in signal processing. They drive the development of algorithms for approaching theoretical results within demanding practical constraints, and provide opportunities for devising clever ways to take advantage of hardware capabilities. State-of-the-art contributions are presented on three topics in this field.The first contribution is the hardware, firmware, and software co-design of a wearable hearing aids research system. The system is open-source, easy to develop for, and much more powerful than traditional hearing aids. Its audio performance matches that of standard hearing aids and it can run custom DSP algorithms in usermode with only 2.4 ms of latency. The system also includes local web-based user control and wearable electrophysiology. The second contribution describes using GPU "Tensor Core" matrix multiply hardware to accelerate the computation of discrete Fourier transforms of sizes which are prime or have large prime factors. This includes mapping these sizes to the power-of-2-size Tensor Cores and emulating higher-precision arithmetic with lower-precision floating point numbers. For large batch sizes and for certain transform sizes which are odd or an odd number times 2 or 4, this approach produced state-of-the-art Fourier transform throughput. Finally, two papers on algorithms design in real-time acoustic modeling for an audio spatialization system are presented. Two perceptually relevant types of diffraction are simulated with ray-based models of sound propagation. Existing methods have accuracy or performance limitations, especially in dynamic applications. A set of algorithms called Volumetric Diffraction and Transmission (VDaT) is introduced to approximate shadowed or near-shadowed diffraction by an occluding object. Similarly, Spatially Sampled Near-Reflective Diffraction (SSNRD) handles near-reflective diffraction involving the edges of reflecting objects. Both methods use ray tracing to spatially sample the scene, approximate ground truth results to within 1-3 dB,and have fast performance suitable for real-time applications. SSNRD also incorporates path generation algorithms, uses a small deep neural network (DNN) to compute the response of each acoustical path, and applies the GPU "RT core" real-time ray tracing hardware to spatial computing tasks beyond traditional ray tracing.

Methods for measuring fish populations are incredibly outdated. Even with rapid improvements in battery technology, compute power and camera systems in the last few years, the most popular techniques for counting fish and measuring their lengths employ humans. In this paper, I present FishSense as an alternative. By utilizing a compact 3-D camera, powerful embedded compute system, and easily-available off-the-shelf parts for assembly, the FishSense modules are sleek, cost-effective and modular systems able to slot into a wide variety of applications. By capturing both depth and color image data, these modules are able to measure the length of individual fish as well, offering an effective non-invasive solution to length \& biomass estimation. The hardware design allows for a high degree of modularity, with options for longer battery life, better performance, and applications in longer-term passive monitoring. The FishSense handheld model has been rigorously tested in several deployments, ranging from a small testing pool with toy fish, to a 70,000 gallon aquarium tank, to an active aquaculture center, with more planned in upcoming months, and preliminary software results from these deployments demonstrate the efficacy of the FishSense modules at measuring fish populations and sizes.

Automatic Plate Recognition plays an important roll in intelligent transportation systems. However, most license plate recognition methods work under restricted conditions like slow speed and good illumination. That is a restriction on industrial application. In this thesis, the constraints are relaxed by vanished points distortion-recovery method and denoising method. This thesis implements a license plate recognition method by morphological edge detection method and convolution neural network recognition method. The thesis is constructed contributes to several papers optimization methods. The proposed approach can be trained for recognition of country-specific license plates. More than 500 images are collected for training and over 300 images are collected for recognition test. This paper achieves 97.05% on license plate recognition for detecting total characters and numbers of the license plates.License plate recognition consists three parts, pre-processing image, locating license plate and identifying license numbers and characters. License plate location is important to obtain license images and plays a key role in identifying plates. The plate recognition has two major steps, character separation and identification. In this paper, machine learning method is applied for license plate recognition.

Modern data science calls for statistical inference algorithms that are both data-efficient and computation-efficient. We design and analyze methods that 1) outperform existing algorithms in the high-dimensional setting; 2) are as simple as possible but efficiently computable. One line of our work aims to bridge different fields and offer unified solutions to multiple questions. Consider the following canonical statistical inference tasks: distribution estimation, functional estimation, and property testing, sharing the model that provides sample access to an unknown discrete distribution. In a recent paper in NeurIPS '19, we showed that a single, simple, and unified estimator – profile maximum likelihood (PML), and its near-linear time computable variant are sample-optimal for estimating multiple distribution attributes. The result covers 1) any appropriately Lipschitz and additively separable functionals; 2) sorted distribution; 3) R\'enyi entropy; 4) $\ell_2$ distance to the uniform distribution, yielding an optimal tester for distributions' closeness. This work makes PML the first unified sample- and time-optimal method for the learning tasks mentioned above. A single algorithm with such broad applicability is \emph{universal}. Another line of our work focuses on instance-optimal learning that designs algorithms with near-optimal guarantees for every possible data input. A flagship problem is distribution estimation over discrete or continuous domains, where ordering and geometry play an essential role. Going beyond worst-case guarantees, researchers designed algorithms that compete with a genie estimator that knows the actual distribution but is reasonably restricted. To obtain state-of-the-art algorithms for both tasks, we leveraged the simple but nontrivial idea of ``data binning''. For discrete settings, we group symbols that appear the same number of times. And for continuous settings, we partition the real domain and separate symbols according to pre-designed local quantiles. The respective algorithms run in near-linear-time, achieve the best-known estimation guarantees regarding the genie estimators, and appear in ICML'19 and NeurIPS '20. A genie-like algorithm adaptive to almost every data sample is \emph{competitive}.We present a comprehensive understanding of universal and competitive algorithms for multiple fundamental learning problems. Our ideas and techniques may shed light on key challenges in modern data science and numerous applications beyond the scope of this dissertation.

Image registration is a major field within computer vision and is often a required step in properly fulfilling other computer vision and pattern recognition tasks such as change detection, scene classification and image segmentation. Recent advances in 3D computer vision and lowered costs in Light Detection and Ranging devices, better known as LiDAR, have given way to an increase in readily available 3D image datasets. These 3D captures give an extra dimension to computer vision data and allow for improvements in a multitude of tasks when compared to their 2D counterparts. However, due to the large scale and complex nature of 3D point cloud data, classical methods for registration often require increased hardware usage and time and can fail to proper register data with a low degree of error. The strategy presented in this paper aims to reduce the number of points representing a point cloud in order to reduce time and hardware overhead needed to perform registration while allowing the algorithm to improve registration accuracy and reduce error between registered clouds. This is done by extracting key edge features from the point clouds using eigenvector analysis to remove ground planes and large normal planes within the point cloud. The algorithm is further improved by performing set differencing on two separate edge extractions to remove large clusters of points representing natural objects that can often cause confusion for registration of outdoor LiDAR scenes. The method for key point registration is evaluated on large scale, complex LiDAR point clouds obtained from aerial sensors. Tests are performed on both fully overlapping and partially overlapping clouds to ensure that the method increases performance on full and partial registration tasks. The tests are also performed on clouds of varying resolution to test the algorithms ability to maintain integrity regardless of cloud resolution. Point reduction results, registration statics and visual results are presented for comparison. A brief look into possible applications of the method and future improvements to the algorithm are included.

Future SAE Level 4 and Level 5 autonomous vehicles will require novel applications of localization, perception, control and artificial intelligence technology in order to offer innovative and disruptive solutions to current mobility problems. Accurate localization is essential for self driving vehicle navigation in GPS inaccessible environments. This thesis concentrates on low speed autonomous shuttles that are mainly utilized for university campus intelligent transportation systems and presents initial results of ongoing work on developing solutions to the localization and perception challenges of a university planned pilot deployment orientated application. The paper treats autonomous driving with real time kinematics GPS (Global Positioning Systems) with an inertial measurement unit (IMU), combined with simultaneous localization and mapping (SLAM) with three-dimensional light detection and ranging (LIDAR) sensor, which provides solutions to scenarios where GPS is not available or a lower cost and hence lower accuracy GPS is desirable. The in-house automated low speed electric vehicle from the Automated Driving Lab is used in experimental evaluation and verification. An improved version of Hector SLAM was implemented on ROS and compared with high resolution GPS aided localization framework in the same hardware architecture. The overall configuration that combines ROS with DSpace controller can be easily transplantable prototype in other hardware architectures for future similar research. Real-world experiments that are reported here have been conducted in a small test area close to the Ohio State University AV pilot test route. are used for demonstrating the feasibility and robustness of this approach to developing and evaluating low speed autonomous shuttle localization and perception algorithms for control and decision making.

This dissertation develops an interdisciplinary approach to integrating renewable energy resources into energy markets, using tools from optimization theory, power systems, and economics. It advances prior work with the development of a set of tools for securing distributed and fully-decentralized optimization problems, including both algorithmic guards against attacks by malicious nodes, and system architectures which can enable decentralized security checks. Leveraging the emerging technologies of blockchains and smart contracts, it develops a new paradigm of blockchain-secured distributed optimization, demonstrated with simulation of a microgrid which is able to securely operate without oversight from a utility or central operator.As the goal of interdisciplinary research is to show mastery of each field and extend current knowledge by exploring their intersection, the chapters of this dissertation are designed to support that goal:• Chapter 1 provides background on the technical challenges of integrating high amounts of renewable energy into the electricity system, and discusses technologies and policies which can address those challenges.• Chapter 2 introduces the idea of applying optimization tools to energy systems by studying a small-scale energy harvesting system in which batteries and capacitors are used to meet a defined load, and constraints force the use of nonlinear optimization techniques.• Chapter 3 explores how large-scale energy storage systems can be designed and sited to maximize profits from participating in wholesale energy markets, using a linear program which demonstrates how convex optimization tools can be united with energy market data to create scalable tools for modeling large networks.• Chapter 4 expands the study of market-based models by examining the strategic operation of generation resources on a congested network. We use game theory to model participants’ behavior, power flow models to reflect the underlying constraints of the physical network, and robust convex optimization to explore how uncertainty is integrated into decision-making.• Chapter 5 discusses how decentralized optimization models can facilitate scalable optimization tools, and explores the security risks which these optimization models introduce. Tools for detection and mitigation of attacks are explored and tested on a simple problem. Potential architectures for securing decentralized optimization are explored, including blockchains and smart contracts.• Chapter 6 extends this approach by developing a blockchain-secured fully-decentralized optimization for a microgrid dispatch problem, coordinating distributed energy resources. By demonstrating the usefulness of this blockchain-secured optimization model, this culminating chapter shows how difficult optimization models can be scalably solved in a manner which respects privacy while guaranteeing security.• The Appendices present tutorial material on the tools used throughout the paper, and are supplemented by the code in the author’s github repository: https://github.com/emunsing/tutorials

Automatic modulation classifiers (AMC) are one of the basic building blocks of electronic warfare receivers and cognitive radios. Although many research papers on AMC algorithms have been published, very few results on their implementation are available. This thesis presents a feature-based AMC built upon a software-defined radio platform. The developed AMC can detect signals over a broad spectrum and classify the modulation used. The modulation schemes considered in this thesis are amplitude modulation (AM), frequency modulation (FM), phase-shift keying (PSK), and quadrature amplitude modulation (QAM). Experimental results demonstrate the validity of the developed AMC algorithm and its implementation.

There are 6.8 million people in the United States that have mobility disorders and must rely on the use of assistive devices to aid in locomotion \cite{kaye2000mobility}. The human body was designed for a particular bipedal locomotion pattern, and deviations from that method of locomotion can result in secondary injuries. Many of the current solutions for mobility disorders are primitive, such as crutches or canes if possible and wheelchairs when not. In recent years, more intelligent, robotic systems have been developed to aid this population that needs assistance walking. Among these new robotic solutions are robotic exoskeletons. Robotic exoskeletons have become a more popular rehabilitation tool in recent years, particularly for those with spinal cord injury. There are currently three robotic exoskeletons with US Food and Drug Administration approval for clinical use. One barrier to wider adoption is the cost of these devices. To reduce cost, a more minimal design was developed at the Human Engineering and Robotics Laboratory at the University of California with only two actuators at the hips and a semi-passive locking mechanism at the knee. While this design can help reduce the weight and cost, it introduces additional complexity to the gait development due to the lack of actuation. The gait development strategies discussed in this paper are inspired by clinical gait data collected from healthy subjects. After all, the ultimate goal of an exoskeleton system as a rehabilitation device is to rehabilitate the pilot to the point that the patient's gait is restored to a natural walking pattern and the device is no longer necessary. Using clinical gait data and biomechanical studies, several models are developed for the design considered. A kinematic model was developed to better understand the angular constraints during both single stance and double stance. A dynamic model was also developed that models the behaviour of the system for different phases of the gait. Finally all these models were linked together in a finite state machine to form a hybrid automaton. The finite state machine specifies the switching conditions for each state. One method for gait development is to design gaits that are tunable such that the gait practitioner can tune the gait to the pilot's comfort and rehabilitation needs. In this endeavour, a kinematic model is used to define the constraints of the double stance phase. The gaits are then generated using a quartic polynomial spline using the node parameters from the kinematic analysis. This method empowers the gait practitioner by using tuning parameters that they understand from gait rehabilitation literature. Another method for gait development is to use optimal gaits based on a hybrid automaton model of the system. The automaton implements three states or phases of the gait cycle including their associated dynamic models. Using this model, the gaits can be optimized for the torques necessary for these gaits under minimal input assumptions from the pilot in each of the phases. This optimization allows for a more detailed understanding of the system dynamics and an optimal gait given constraints of both the system and the gait cycle. Furthermore, the optimization method can be utilized as a gait generator that is tailored to each individual pilot, effectively reducing the workload of the gait practitioner.

Vehicle classification has become a very important subject of study because of its importance in autonomous navigation, surveillance and traffic analysis. Classification of vehicles from the rear view is challenging because all the vehicles have subtle appearance differences from the rear view, changing illumination conditions, presence of shadows and real-time considerations. While numerous approaches have been introduced for this purpose, no specific study has been conducted to provide a robust and complete video-based vehicle classification system based on the rear view where the camera’s field of view is directly behind the vehicle. In this paper we present a multi-class vehicle classification system which classifies a vehicle into one of four possible classes Sedan, Minivan, SUV and Pickup truck when seen from its rear view. For a given geometric setup of the camera we use a feature set of Visual Rear Ground Clearance, height of the vehicle and perpendicular distance between the bottom of the license plate and bottom of the rear bumper for classifying the vehicle. Results are shown on large data-sets of freeway videos.

SQL-on-Hadoop systems, query optimization, data distribution over multiple nodes and parallelization techniques are few of the areas under extreme research these days. Big names like Amazon, Google, Microsoft and many more are working on implementing systems for faster access of data from multiple nodes, reducing data mobility and increasing the parallelization. Customer’s queries are retrieved and reviewed by the database systems in an efficient way in the least amount of time by the introduction of various parallelization techniques by running the same query in parallel over different nodes carrying the data. Apart from multi-threading parallelization, there is another way of parallelization that can be performed in order to further reduce retrieval time, hence improving the efficiency of the system; parallelization on user queries on top of a DBMS/RDBMS. In this paper, we will study one such technique of how multiple queries can run simultaneously on a system in order to increase the system efficiency by reducing the data retrieval from the storage. Maximum sharing of workload has been performed by generating optimal and ubiquitous join plans for a set of queries and then fed them to GLADE (Generalized Linear Aggregate Distributed Engine), a scalable distributed system for large scale data analytics. Our main work is centered on generating GLADE join plans for a Multi-Query, satisfying maximum number of queries in order to maximize data sharing and minimize data retrieval for each individual query.

With the advent of electronic health records, more data is continuously collected for individual patients and more data is available for review from past patients. Despite this, it has not yet been possible to successfully use this data to systematically build clinical decision support systems that can produce personalized clinical recommendations to assist clinicians in providing individualized healthcare. In this paper, we present a novel approach, Discovery Engine (DE) that discovers which patient characteristics are most relevant for predicting the correct diagnosis and/or recommending the best treatment regimen for each patient. We demonstrate the performance of DE in two clinical settings: diagnosis of breast cancer as well as personalized recommendation for a specific chemotherapy regimen for breast cancer patients. For each distinct clinical recommendation, different patient features are relevant; DE can discover these different relevant features and use them to recommend personalized clinical decisions. The DE approach achieves a 16.6% improvement over existing state-of-the-art recommendation algorithms in terms of kappa coefficients for recommending the personalized chemotherapy regimens. For diagnostic predictions, the DE approach achieves a 2.18% and 4.20% improvement over existing state-of-the-art prediction algorithms in terms of prediction error rate and false positive rate, respectively. We also demonstrate that the performance of our approach is robust against missing information, and that the relevant features discovered by DE are confirmed by clinical references.

Face recognition has become more significant in recent years. Facial landmarks, on the other hand, have been proven to be useful in related fields such as face recognition and facial expression analysis. In this paper, we propose a method to identify frontal view faces using facial landmarks, and show that it is robust to 2D rotation, change of scale and change of position. We propose a face descriptor revealing the shape features, the Euclidean distance between landmarks and their mean. We evaluate the effectiveness of the representation by using it to perform face classification on our emotion dataset. We fit a multivariate normal distribution for each identity and construct a Naive Bayes Classifier for classification. We show that our method performs well with a small classification error.

In order to deal with recognition problem of the low-resolution and poor-quality images captured from traffic surveillance camera in intelligent transportation system, a vehicle logo super-resolution method using two-dimensional canonical correlation analysis(2D CCA)is presented. While numerous approaches have been proposed to do the recognition, they can only achieve good performance on the images with high resolution. In this paper, a novel combination of morphological operation and sliding window technique is adopted to detect the vehicle logo, and then using 2D CCA to do super-resolution reconstruction which improve performance significantly compared with other traditional methods. It analysis the performance of proposed approach on a large image dataset.

The main challenge in designing the future heterogeneous many-core architecture on the same chip is to provide a solution that has low power consumption, in addition to trade off a small decrease in performance and throughput. Our design incorporates heterogeneous cores representing different points in the power-performance design space during an applications execution. Under this circumstance, system software dynamically chooses the most appropriate core to meet specific performance and power requirements. In this paper, the authors present a power-aware core management scheme based on efficient control of the core resources on heterogeneous many-core architecture as a mechanism to reduce a huge latency and a power dissipation for powering the core up from powered down. Operation is based on distinct scenarios by 3-bit core power control scheme through 5 statuses switching such as active, hot core, cold core, idle, and powered down. In addition, for more elaborated control to be power-performance efficient, this kind of status switching is exactly triggered by power-aware thread placement through heuristic thread consolidation approach. To achieve this objective, we have tried to deal with this scheme in terms of calculating peak and typical power consumption and managing core resources efficiently.

This paper examines the combination of adaptive signal design and orthogonal frequency division multiplexed (OFDM) pulses in order to improve their detection performance and automatic target recognition for multiple varieties of single target scenarios. This adaptive design will be performed through exploiting the target’s reflectivity function. OFDM’s performance for radar systems is discussed and analyzed, as well as its potential in adaptive design. The place of adaptive signal design in developing radar systems is examined, and methods are detailed. Estimators for determining the target’s reflectivity function (TRF) are derived and implemented. Finally, performance results for TRF identification and automatic target recognition are discussed.

Modern radar imaging requires advanced Synthetic Aperture Radar (SAR) techniques in order to compensate for the low resolution typically found on airborne radars. However, the techniques used to process the data collected during SAR operation require a great deal of computation. This paper proposes a novel efficient algorithm and approach to processing the data collected to quickly get high-resolution approximations for the locations of non-moving targets.

Optical tomosynthetic imaging of 3-D objects from disparate 2-D images has been limited in the past by a lack of algorithmic enforcement of physical constraints, in particular within-scene obscuration and object self-occlusion. This paper presents a stochastic observation model of a tomosynthetic collection that explicitly includes an obscuration operator that is unknown by the sensor. The expectation-maximization algorithm is used to iteratively estimate the obscuration operator and to reconstruct the 3-D volume of interest. Explicit inclusion of obscuration effects greatly enhances the spatial and spectral accuracy of 3-D results without use of costly post-processing techniques. Performance metrics are introduced and the resultant receiver operating characteristics are presented.

Standard procedures used by law enforcement for evidence preservation require removing power from the computer and going further analysis in a digital forensics laboratory. With the advent of easy to use, highly accessible hard drive encryption, access to potential evidence becomes much more volatile problem than in the past. This paper will look at several encryption methods available on the commercial market, discuss their impact on current best practices for hard drive recovery, and propose new procedures to deal with the possibility of encrypted information on a target computer.

This paper presents a concept on how the software requirements specifications template provided by IEEE Standard 830 could be updated to ensure that security is analyzed during the early stages of the software development lifecycle. This improved security requirement in the software requirements specifications will ensure that software developers will have a more clear understanding of how to protect digital information.