教学课程资源库 更多>>

关键词： Artificial intelligence   Computer science  

摘要： Over the past 20 years, data has increased on a large scale in various fields and researchers are facing new challenges to work with this huge volume of data. My passion for high-performance computing leads me to span my research across different fields in computer science. In this dissertation, we introduced high-performance solutions to problems in machine learning and bioinformatics like crater detection on Mars images and structural variation detection on DNA/RNA genome data. We first show different parallel heterogeneous implementations of the Berlekamp-Massey Algorithm which plays an important role in error correction of NAND flash memories which can be found almost everywhere from flash drives to large scale enterprise servers. Then, we introduce a semi-supervised learning technique to learn a novel weighted-distance metric named WDM which learns labeled information from the training set and identifies groups among the samples from the test set to form a metric space. The effectiveness of this novel metric evaluated intensively on classification tasks and on both CPU and GPU parallel platforms. This dissertation continues with the introduction of a bidirectional context-based deep learning framework that learns bidirectional context-based features from both craters and its surrounding features using deep convolutional classification and segmentation models to identify efficiently sub-kilometer craters in high-resolution panchromatic images. Finally, we designed two computational methods to detect cancer inter-chromosomal rearrangements and gene fusions on DNA and RNA sequencing data, respectively. These methods split the candidate reads into windows and then represent it using the binary format to reduce the huge dimension of the search space and then search the location of the breakpoint in the reference genome using Jaccard distance. This dissertation, in its entirety, offers efficient solutions to a few problems in coding theory, machine learning, and bioi

关键词： distributed storage systems   privacy-preserving systems   Computer science   Computer science and engineering  

摘要： Today's mobile devices sense, collect, and store huge amounts of personal information, which users share with family and friends through a wide range of cloud applications. Many of these applications are large-scale – they must support millions of users from all over the world. These applications must manage a massive amount of user-generated content, they must ensure that this content is always available, and they must protect users' privacy. To deal with the complexity of managing a large amount of data and ensure its availability, large-scale cloud applications rely on transactional, fault-tolerant, distributed storage systems. Latest hardware advancements and new design trends, such as in-memory storage, kernel-bypass message processing, and geo-distribution, have significantly improved the transaction processing times. However, they have also exposed overheads in the protocols used to implement these systems, which hinder their performance; specifically, existing protocols require unnecessary cross-datacenter and cross-processor coordination. This thesis introduces Meerkat, a new replicated storage system that avoids all cross-core and cross-replica coordination for transactions that do not conflict. Moreover, replica recovery protocols that do not require persistent storage devices, are proposed; these protocols can be easily integrated with existing storage systems. Almost all applications offer their users a choice of privacy policies; unfortunately, they frequently violate these policies due to bugs or other reasons. Therefore, this thesis introduces SAFE, a privacy-preserving system that automatically enforces users' privacy policies on untrusted mobile-cloud applications. Our results demonstrate that SAFE is a practical way for users to create a chain of trust from their mobile devices to the cloud.

关键词： Computer science  

摘要： Instrumentation is fundamental to developing and maintaining applications in the modern datacenter. It affords visibility into what an application is doing at runtime, and it helps pin-point bugs in a system by exposing the steps that lead to an error. The most common method of instrumentation today is logging, or printing out human-readable messages during an application's execution. Unfortunately, as applications have evolved to become increasingly more performant with tighter latency requirements, traditional logging systems have not kept up. As a result, the cost of producing human-readable log messages is becoming prohibitively expensive. NanoLog is a nanosecond scale logging system that's 1-2 orders of magnitude faster than existing logging systems such as Log4j2, spdlog, Boost log, or Event Tracing for Windows. The system achieves a throughput of up to 82 million log messages per second for simple log messages and has a typical log invocation overhead of 8 nanoseconds. For comparison, other modern logging applications today can only achieve up to a few million log messages per second at log latencies of hundreds of nanoseconds to several microseconds. NanoLog achieves its ultra-low latency and high throughput by shifting work out of the runtime hot-path and into the compilation and post-execution phases of the application. More specifically, it performs compile-time extraction of static information from the log messages to reduce I/O and decouples formatting of the log messages from the runtime application by deferring it until after execution. The result is an optimized runtime that only outputs the minimal amount of data and produces a compact, binary log file. The binary log file is also amenable to log analytics engines; it is small relative to full, human-readable log messages and contains all the data in a binary format, saving the engine from parsing ASCII text. With these enhancements, NanoLog enables nanosecond scale logging and hopes to fill the per

关键词： Information technology   Computer science  

摘要： The necessities of modern day lives today depend heavily on usage of the Internet of Things (IoT). Exponential growth of this technology has led to increase in security and privacy concerns. Heterogeneity of the IoT Sensors is a major limitation for security decision makers who have to extract and integrate data from different IoT sensors to assess overall security posture of the network. Current security solutions cater to homogeneous networks where devices use similar configurations and protocols for communication. In this work we describe the IoT Message Abstraction Format (IoTMAF), which is a first step towards IoT data standardization. IoTMAF is capable of abstracting wireless IoT protocol data without losing any of its significance. We demonstrate that IoTMAF aids in the detection of a complex attack targeting a fully heterogeneous environment. In addition, we describe the eXploitable IoT ecosystem (xIoTec), which is intended to be used for heterogeneous IoT security research. xIoTec is a fully heterogeneous IoT security testbed. The testbed includes both network and physical attack vectors. Moreover, it implements the full IoT ecosystem taking account of user interaction with IoT devices and cloud services. We were able to conduct a security research experiment using this testbed by performing network and physical attacks using the IoT devices.

关键词： Computer science  

摘要： Theoretical computer science is an essential part of computer science study programmes. Many articles from different countries can be found, which are concerned with the high levels of frustration amongst students, motivational problems, and high failure rates in courses on theoretical computer science. There are even indications that theoretical computer science is one of the reasons for the high dropout rates of computer science study programmes. This dissertation presents a design-based research approach, aiming to improve the learning in university courses on theoretical computer science. For this approach, the author created interactive Learning Units in the learning management system Moodle. As theoretical basis, the cognitive load theory, the cognitive theory of multimedia learning and the cognitive-affective theory of learning with media as well as the Felder-Silverman model of learning styles were analyzed. All these theories were used to derive design recommendations for the Learning Units. Furthermore, existing approaches concerning e-learning in the area of theoretical computer science and meta-analyses on e-learning and blended learning in general were taken into account. The created Learning Units use text and video as presentation forms and exercises on the content. Students can follow different learning paths on the content. On these paths, they can choose whether they want to be presented with simple examples introducing a new subject before its actual explanation. Another possible choice at several points is to follow a further extension on the content of the explanation if they are more deeply interested. These Learning Units were intended for self-studying and self-testing using exercises. One pair of Learning Units were created for a course on formal languages and automata, typically attended at the beginning of the study programme, and another pair for a more advanced course on reactive systems. In both cases, the Learning Units were intended a

关键词： Computer science   Diversity   Fairness   Machine Learning   Multi-armed bandits  

摘要： With algorithms, artificial intelligence, and machine learning becoming ubiquitous in our society, we need to start thinking about the implications and ethical concerns of new machine learning models. In fact, two types of biases that impact machine learning models are social injustice bias (bias created by society) and measurement bias (bias created by unbalanced sampling). Biases against groups of individuals found in machine learning models can be mitigated through the use of diversity and fairness constraints. This dissertation introduces models to help humans make decisions by enforcing diversity and fairness constraints. This work starts with a call to action. Bias is rife in hiring, and since algorithms are being used in multiple companies to filter applicants, we need to pay special attention to this application. Inspired by this hiring application, I introduce new multi-armed bandit frameworks to help assign human resources in the hiring process while enforcing diversity through a submodular utility function. These frameworks increase diversity while using less resources compared to original admission decisions of the Computer Science graduate program at the University of Maryland. Moving outside of hiring I present a contextual multi-armed bandit algorithm that enforces group fairness by learning a societal bias term and correcting for it. This algorithm is tested on two real world datasets and shows marked improvement over other in-use algorithms. Additionally I take a look at fairness in traditional machine learning domain adaptation. I provide the first theoretical analysis of this setting and test the resulting model on two deal world datasets. Finally I explore extensions to my core work, delving into suicidality, comprehension of fairness definitions, and student evaluations.

关键词： Artificial intelligence   Computer science  

摘要： Machine learning systems have been shown to be vulnerable to adversarial examples. We study the most practical problem setup for evaluating adversarial robustness of a machine learning system with limited access: the hard-label black-box attack setting for generating adversarial examples, where limited model queries are allowed and only the decision is provided to a queried data input. Several algorithms have been proposed for this problem but they typically require huge amount (>20,000) of queries for attacking one example. Among them, one of the state-of-the-art approaches (Cheng et al., 2019) showed that hard-label attack can be modeled as an optimization problem where the objective function can be evaluated by binary search with additional model queries, thereby a zeroth order optimization algorithm can be applied. In this thesis, we adopt the same optimization formulation but propose to directly estimate the sign of gradient at any direction instead of the gradient itself, which enjoys the benefit of single query. Using this single query oracle for retrieving sign of directional derivative, we develop a novel query-efficient Sign-OPT approach for hard-label black-box attack. We provide a convergence analysis of the new algorithm and conduct experiments on several models on MNIST, CIFAR-10 and ImageNet. We find that Sign-OPT attack consistently requires 5X to 10X fewer queries when compared to the current state-of-the-art approaches and usually converges to an adversarial example with smaller perturbation.

关键词： Computer science   Robotics   3D Learning   Computer Vision   Differential Cameras   Event-Based Sensors   Motion Estimation   Segmentation  

摘要： Computer vision has been dominated by classical, CMOS frame-based imaging sensors for many years. Yet, motion is not well represented in classical cameras and vision techniques - a consequence of traditional vision being frame-based and only existing 'in the moment' while motion is a continuous entity. With the introduction of neuromorphic hardware, such as the event-based cameras, we are ready to cross the bridge of frame based vision and develop a new concept - motion-based vision. The event-based sensor provides dense temporal information about changes on the scene - it can ‘see’ the motion at an equivalent of almost infinite framerate, making a perfect fit for creating dense, long term motion trajectories and allowing for a significantly more efficient, generic and at the same time accurate motion perception. By its design, an event-based sensor accommodates a large dynamic range, provides high temporal resolution and low latency -- ideal properties for applications where high quality motion estimation and tolerance towards challenging lighting conditions are desirable. The price for these properties is indeed heavy - event-based sensors produce a lot of noise, their resolution is relatively low and their data - typically referred to as event cloud - is asynchronous and sparse. Event sensors offer new opportunities for robust visual perception so much needed in autonomous robotics, but challenges associated with the sensor output, such as high noise, relatively low spatial resolution and sparsity, ask for different visual processing approaches. In this dissertation we develop methods and frameworks for motion segmentation and egomotion estimation on event-based data, starting with a simple optimization-based approach for camera motion compensation and object tracking and continuing by developing several deep learning pipelines, while continuing to explore the connection between the shapes of the event clouds and scene motion. We collect EV-IMO - the first pixelw

关键词： Information science   Computer science  

摘要： Threats posed by malware have thrust cybersecurity to the forefront of system design considerations. Domain-specific constraints prohibit many devices from supporting traditional host-based protections, which themselves can be subverted by advanced malware. This research proposes a platform agnostic anomaly-based intrusion detection system (ADS) that operates outside malware’s sphere of influence. The ADS resides on-chip and collects thermal side-channels via device-level sensors implemented alongside an embedded processor. Changes in this side-channel correlate to global system behavior, providing an unsubvertible intrusion detection source. We investigate the efficacy of this approach by evaluating the accuracy of traditional classification algorithms for datasets collected from benign and infected Linux environments on the target device. We then explore an additional use case for this side-channel called divergence detection. Our experimental methodology seeks to determine to what extent are device-level thermal side channels effective as an intrusion detection source for constrained systems?