教学课程资源库 更多>>

关键词： Computer science   Street signs   Neural networks   Probability   Methods   Algorithms   Probability distribution   Competition   Artificial intelligence  

摘要： This dissertation explores the problem of robust classification against adversarial examples. While modern classification algorithms achieve high accuracy on innocuous samples, they show significantly lower performances after adding imperceptibly small perturbations, so-called adversarial perturbations, to those innocuous samples. Designing a robust classification against such adversarial examples is a crucial topic in machine learning community to make machine learning algorithms applicable in real-world situations. To deeply understand adversarial examples, this dissertation delves into two different aspects about this topic—attack side and defense side. On the attack side, we focus on attack algorithms generating adversarial examples for a given classifier. To understand how adversarial perturbations are generated, we design a new gradient descent algorithm—an algorithm that uses the gradient of a given classifier with respect to the input. Based on Newton's method, our algorithm achieves quicker convergences to adversarial points. Also, to reflect human imperceptibility, we propose a new set of metrics that are based on contemporary computer vision techniques. By measuring the performance of various attacks with these metrics, we empirically demonstrate that our new algorithm is comparably effective in fooling a classification algorithm. On the first part of the defense side, our main goal is to improve the robustness of classification algorithms by reinforcing existing defense strategies. We first start from a well-known formulation of adversarial training by Madry et al. (2017). From a thorough analysis, we figure out that, under a various instantiations the formulation, the confidence of a classifier on its prediction can be exploited as a discriminator between correct classifications and wrong classifications. Based on this result, we propose a framework that reinforces the adversarial robustness of a given base classifier and experimentally estimate the per

关键词： Cancer   DNA methylation   Gene expression   Bioinformatics   Lasers   Binding sites   Chromosomes   Genomes   Epigenetics   Batch processing   Algorithms   Genomics   Cell division   Biochemistry   Cellular biology   Chemistry   Computer science   Genetics   Optics  

摘要： Third-generation long-read technologies denote the latest progression in high throughput DNA and RNA sequence analysis. Complementing the widespread second-generation short-read platforms, long-read sequencing adds unique application opportunities by generating previously unattainable read lengths. Despite the remaining higher error rate compared to short reads, single-molecule real-time sequencing (SMRT) and nanopore sequencing advanced to be state-of-the-art for de-novo genome assemblies and identification of structural variants. Continuous throughput and accuracy improvements lead to development of novel methods and applications at a fast pace. We identify major application fields and key bioinformatic software for long-read sequencing data analysis by employing a data driven literature research. The integration of citations and keywords into a literature graph provides a scaling approach to analyze an exponentially growing number of third-generation sequencing related publications. Even though sparking the development of countless bioinformatics software, the streamlined nanopore data processing into standardized formats is still lacking. As an enabling step for its successful application, we developed Nanopype, a modular and scalable pipeline. Our approach facilitates the basic steps of basecalling, alignment, methylation- and structural variant detection with exchangeable tools in each module. Optimized for the usage on high performance compute clusters, we propose a raw data management, capable of handling multiple sequencing devices placed locally and remotely. Strict version control of integrated tools and deployment as containerized software, ensure reproducibility across projects and laboratories. Finally, we analyze disease associated repeat regions utilizing targeted nanopore sequencing and the Nanopype processing infrastructure. The expansion of unstable genomic short tandem repeats (STRs) is of particular interest as it causes more than 30 Mendelian h

关键词： Biostatistics   Medical imaging   Computer science   Artificial intelligence  

摘要： Idiopathic pulmonary fibrosis (IPF) is one type of interstitial lung disease (ILD) of unknown causes. High-resolution computed tomography (HRCT) scans play a crucial role in distinguishing IPF from non-IPF among subjects with ILD. This radiological evaluation is an important yet difficult task. In clinical practice, making a correct and reliable IPF diagnosis is critical to ensure patients with different causes of pulmonary fibrosis be treated appropriately and patients with IPF be assessed for novel therapies and lung transplantation. Therefore, this dissertation aims to build an automated IPF diagnosis system for ILD subjects, using volumetric and non-contrast chest HRCT scans. Supervised learning methods are a type of machine learning methods that require labels of training samples as ground truth to learn a mapping function between input and output labels. Depending on the type of labels provided, if only coarse-level labels are available, rather than fine-scale labels, this task is called a weakly supervised task. For our example, acquiring fine-scale information of certain radiological patterns can be helpful for building the diagnostic system, but fine-scale labels are expensive to obtain. On the other hand, coarse-level labels are usually easier to acquire, such as CT scan-level information. Since we only have labels at a CT scan level, our problem is a weakly supervised task. To tackle this challenge, this dissertation leverages domain knowledge acquired from previous studies, including IPF progression and quantification information, to provide more efficient, reliable, and explainable diagnostic support. In project I, we used 2D deep learning models with IPF progression information and optimal design criterion to weigh HRCT samples differently. In project II, 3D deep learning models with multi-scale attention models were used with IPF quantification maps to achieve good model accuracy and explainability. Furthermore, we evaluated the robustness of these de

关键词： Kinematics   Cameras   Magnetic fields   Orbits   Sensors   Meteors & meteorites   Algorithms   Magnetism   Satellites   Aerospace engineering   Astronomy   Computer science   Electromagnetics   Engineering   Mechanics   Physics   Planetology  

摘要： The Meteorix mission is a nanosatellite demonstration mission dedicated to the observation and characterization of meteors and space debris entering into the Earth's atmosphere. The scientific objectives of this mission require a high performance attitude determination and control system (ADCS). This thesis entails the development of attitude estimation and control algorithms for the Meteorix 3U cubesat. Attitude estimation and control requirements are outlined, given the mission's scientific objectives and the power needs of the satellite. A magnetometer-only multiplicative extended Kalmanfilter (MEKF) is formulated to estimate the satellite's attitude and angular velocity. Itis compared to a conventional gyro-based MEKF and uses the derivative of the magnetic field and the attitude dynamics model, in place of gyroscope measurements to update the angular velocity estimate. Results from Monte Carlo simulations show that the magnetometer-only MEKF outperforms a gyro-based MEKF in low angular velocity scenarios. A sliding mode controller is developed for attitude control using three magnetic actuators, while a single flywheel oriented in a momentum bias configuration provides gyroscopic stability. Using the attitude and angular velocity feedback from the magnetometer-only MEKF, the sliding mode control law offers better pointing accuracy than a standard proportional derivative (PD) magnetic feedback controller; it is also shown to be more robust to uncertainties and disturbances in a sensitivity study. The magnetometer-only MEKF suffers in the presence of model uncertainties such as flywheel misalignment and residual magnetic moment, and methods to improve the overall ADCSperformance are discussed.

关键词： Computer science   Educational technology   Artificial intelligence  

关键词： Computer science   Mathematics   Operations research   Wood sciences  

摘要： This thesis focuses on two fundamental problems in modern industry: ProductionPlanning problem in Supply Chain System and Optimal Matching Problem in Ride-hailingSystem. In the first part, we propose an efficient and flexible decomposition framework forsolving large-scale production planning and scheduling problem, which essentially is ahuge mixed-integer programming problem. We focus on several levels of decomposition:from the model perspective, we filter necessary and valid variables and constraints outof the original problem before construction; for items with similar BOM structures, wedivide them into groups based on spectral clustering analysis and generate a series ofsubproblems; we calculate the priority order for these problems and progressively denote two subproblems with current highest priority as Master problem and Auxiliary problem,which are solved in a parallel fashion; to decompose the Master Problem from thetime perspective, we combine LP relaxation and rolling horizon strategy output a near-optimalsolution. The freedom to choose filtration intensity, subproblems numbers, andthe truncated time window length makes our decomposition algorithm flexible and ***, the gap between the optimal solution to the original problem and theone to our decomposition algorithm is small, and the overall construction and solvingtime is significantly shortened. In the second part, we study the optimal threshold matching policy in a three-stageflexible queueing model for a spatial ride-hailing system with two types of passengers anddrivers, aiming to maximize the throughput and profit for the system. We first get accurateperformance evaluations from a fluid approximation and obtain optimal solutions basedon the unique equilibrium of the fluid process. Furthermore, we design an adaptive profitmaximization algorithm based on the optimality condition to solve problems with time-varyingarrival rates. Simulations provide convincing verification of the near-opt

关键词： Eigen values   Decomposition   Energy   Algorithms   Boundary conditions   Altitude   Computer science   Geography   Mathematics  

摘要： Identification of all physical mechanisms responsible for the flow breakdown from laminar to turbulent remains elusive in many high-speed boundary layer flows. Predicting the onset of transition and understanding the linear mechanisms responsible for transition in supersonic and hypersonic is critical to controlling instability mechanisms and preventing a multifold increase in the thermal protection system required. In the traditional modal theory, low amplitude external disturbances are generated by receptivity mechanisms and grow exponentially, leading the flow to transition. On the other hand, for flows subjected to high external disturbance levels, a bypass of the modal amplification path can happen and a nonmodal stability analysis becomes an additional route to transition that needs to be studied. The present thesis aims to address both modal and nonmodal stability analysis of highspeed spatially-inhomogeneous boundary-layer flows. In this effort, a massively parallel code, Linear Global instability for Hypersonic Transition (LiGHT) has been developed to solve multi-dimensional complex non-symmetric eigenvalue problems (EVP) and Singular Value Decomposition (SVD) problems arising in the solutions of global linear fluid flow instability. The code has been applied to four canonical flow configurations. The first flow studied has been the main motivator of the present thesis: it concerns hypersonic flow over the HiFIRE-5 elliptic cone model, which is investigated at two different flight altitudes. A new physical mechanism, unveiled by transient growth analysis, revealed that optimal conditions are associated with streamwise velocity streaks in the crossflow region. These streaks are in good qualitatively agreement with experimental results. Moreover, a quartic dependence of the maximum energy gain on the local Reynolds number was found in the highest altitude analysed, contrasting with the quadratic dependence found at the lower altitude and known from the incomp

关键词： Energy   Computer science   Alternative energy   Near Eastern studies   Civil engineering   Sustainability   Climate change  

关键词： Computer Science   Computer Security   Concept Inventory   Cybersecurity   Education   Misconceptions  

摘要： Cybersecurity incidents are on the rise. Tracing these security breaches back, we linked them to people making an error due to a commonsense misconception. There is no one standard tool that gauges a student's understanding of security topics. In this research, we surveyed 75 security experts about security novices' misconceptions, coded the results, and identified 17 top misconceptions. We created open-ended questions and labs/active learning to identify and remediate those misconceptions. After revising the open-ended questions, we gave them to undergraduate students and successfully extracted real-world instances of the misconceptions in practice. We created a ten-question multiple-choice exam by converting the open-ended questions into multiple-choice with many distractors drawn from students' misconceptions. We then conducted "think-aloud interviews" with students to make sure that the questions were clear. After integrating their feedback, we administered multiple-choice exams to two groups of students; 114 CS 1 students with no formal security education and 28 students from a security course. Almost 30% of CS 1 students failed to answer more than one question correctly, and only 3.5% of CS 1 students passed with a score of 60 (a D-). However, only 21.4% of the security students passed, and no individual student got more than seven out of ten correct. Our results show that both groups of students have these common security misconceptions. While security students earned markedly higher scores, our test unequivocally shows that students are leaving the security course retaining significant misconceptions, pointing the way for improvements in teaching.

关键词： Electrical engineering   Artificial intelligence   Computer science