教学课程资源库 更多>>

关键词： Computer science   Computer engineering   Technical communication   Biochemistry  

摘要： In cybersecurity, understanding the technologies and the best ways to interface with them is paramount for staying ahead of growing cyberthreats. Developers of cybersecurity software will benefit greatly from a greater understanding of how users prefer to interact with cybersecurity technology. In the modern world, two primary interface methods are currently used: the command-line interface (CLI) and the graphical user interface (GUI). This study is a survey and introspective into what benefits and drawbacks that each method has when in the hands of users who do not have a comprehensive background in cybersecurity. Untrained individuals showed proficiency when working with GUI systems, showing that developing modern cybersecurity systems with GUIs would improve ease of use for such individuals. Additionally, the CLI was favorable for more complex operations but was difficult for users who were not accustomed to the CLI.

关键词： Computer science   Tomography   Datasets   Brain cancer   Magnetic resonance imaging   Neurosciences   Medical imaging   Information technology  

关键词： Computer science   Artificial intelligence   Computer engineering  

摘要： While capable of achieving state-of-the-art performance in complex sequential tasks, deep reinforcement learning (deep RL) remains extremely data inefficient and slow to train. This slow learning speed poses challenges for applying deep RL to real-world situations, especially when poor initial performance is unacceptable or even dangerous. Many approaches have been studied to tackle this problem and one widely used method is transfer learning (TL). The principle of TL is that knowledge acquired from a source agent can be leveraged to assist learning in a different but related target task. This dissertation proposes three types of TL techniques to speed up the learning of a deep RL agent. Specifically, we demonstrate that knowledge can be transferred agent-to-agent, human-to-agent, and self-to-agent. First, we show that positive transfer can be achieved between two cross-domain agents via direct weight copying if they share visual similarities. Second, we study various pre-training methods using a set of human demonstrations to perform the human-to-agent transfer. Pre-training significantly speeds up the agent's learning. Third, we explore knowledge transfer from the agent to itself via a novel experience replay framework, namely Lucid Dreaming for Experience Replay (LiDER), in which past experiences are constantly refreshed. Results suggest that the agent can achieve much better performance within the same amount of training data compared to the case without replaying refreshed experiences. Two extensions of the LiDER framework also enable agent-to-agent and human-to-agent transfer, making it a powerful tool to perform all three types of transfer.

关键词： Software engineering   Sensors   Engineering   Computer science   Automotive engineering  

关键词： Software   Manufacturing   Success   Optimization   Robots   Linear programming   Algorithms   Traveling salesman problem   Cities   Computer aided engineering--CAE   Engineers   Robotics   Computer science   Design   Industrial engineering   Operations research  

摘要： In industry, short ramp-up times, product quality, product customization and high production rates are among the main drivers of technological progress. This is especially true for automotive manufacturers whose market is very competitive, constantly pushing for new solutions. In this industry, many of the processes are carried out by robots: for example, operations such as stud/spot welding, sealing, painting and inspection. Besides higher production rates, the improvement of these processes is important from a sustainability perspective, since an optimized equipment utilization may be achieved, in terms of resources used, including such things as robots, energy, and physical prototyping. The achievements of such goals may, nowadays, be reached also thanks to virtual methods, which make modeling, simulation and optimization of industrial processes possible. The work in this thesis may be positioned in this area and focuses on virtual product and production development for throughput improvement of robotics processes in the automotive industry. Specifically, the thesis presents methods, algorithms and tools to avoid collisions and minimize cycle time in multi-robot stations. It starts with an overview of the problem, providing insights into the relationship between the volumes shared by the robots' workspaces and more abstract modeling spaces. It then describes a computational method for minimizing cycle time when robot paths are geometrically fixed and only velocity tuning is allowed to avoid collisions. Additional requirements are considered for running these solutions in industrial setups, specifically the time delays introduced when stopping robots to exchange information with a programmable logic controller (PLC). A post-processing step is suggested, with algorithms taking into account these practical constraints. When no communication at all with the PLC is highly desirable, a method of providing such programs is described to give completely separated robot wo

关键词： Computer science   Bioinformatics   Computer engineering   Artificial intelligence  

摘要： The multi-way associations across multiple biological networks carry rich information of the dependencies between heterogeneous biological objects such as human diseases, genes (proteins) and drugs. Inferring this information based on the network topologies has emerged as a core step in many bioinformatics tasks. Due to the modeling and computational challenges, current studies mainly focus on predicting the bipartite associations across pairs of biological networks rather than predicting the multi-way associations; and most of the existing multi-relational learning methods are not scalable for predicting high-order associations across a large number of networks. The goal of our research is to advance the field of multi-relational learning by systematically addressing the modeling and computational challenges. To that end, we develop machine learning methods to directly model the multi-way associations across multiple biological networks as a tensor (multi-way array), which is structured by the manifolds in the product of multiple networks; and improve the learning scalability through optimally estimating the tensor and the product graph. These methods rely on tensor decomposition and graph-based transductive learning technologies. First, we propose the Graph-Regularized Tensor Completion from Observed Pairwise Relations (GT-COPR) method to directly infer the multi-way associations among the entities across multiple biological networks in a low-rank tensor using the observed bipartite associations. We validate that compared to the state-of-the-art bipartite relational learning methods, the tensor formulation enables GT-COPR to identify more important pharmacogenomic multi-way associations across disease, gene and chemical networks. Next, we present the Low-Rank Tensor-based Label Propagation (LowrankTLP) method to address scalability challenges in multi-relational learning, by providing a theoretically justified framework to estimate the product graph with respectiv

关键词： Bioinformatics   Computer science   Genetics  

摘要： Due to the growing need for privacy-preserving and secure genome analysis, various tools and techniques have been proposed for the development of bioinformatics software such as Fully Homomorphic Encryption (FHE), Secure Multiparty Computation (SMC), and Trusted Execution Environments (TEE). In response to this, we explored the utility of TEEs, specifically Intel Software Guard eXtensions (SGX) due to its availability, as a solution to core bioinformatics and genome analysis tasks by providing a general framework for privacy-preserving and secure algorithm development. Throughout this process, we designed and implemented three algorithms: SkSES for secure and privacy-preserving Genome-Wide Association Studies(GWAS), ScNN for privacy-preserving machine learning as a service for disease outcome prediction, and SMac for secure genotype imputation. Through our experiments, we demonstrated that we were able to achieve computationally feasible, privacy-preserving, and secure solutions to the three core bioinformatics tasks mentioned above. With SkSES and ScNN, we showed that despite the constantly increasing genomic data volume and hardware resource limitations, it is possible to build scalable solutions using TEEs with the utilization of sketching algorithms and efficient data compression schemes. With SMac, we provide a system that offers excellent runtime performance and protection against all known side-channel attacks against Intel SGXwhile preserving accuracy compared to state-of-the-art genotype imputation software. Ourwork provides an important step towards practical and secure collaborative analysis on the public cloud while protecting patient privacy which could help researchers across the globe share data, build more powerful models, and accelerate biological innovations.

关键词： Strain gauges   Adhesive bonding   Failure   Physics   Wavelet transforms   Sensors   Crack initiation   Aluminum   Algorithms   Kalman filters   Fracture mechanics   Markov analysis   Crack propagation   Computer science   Civil engineering  

摘要： Adhesively bonded joints have been widely accepted and increasingly used in major load-carrying structural components due to many advantages over classical mechanical fasteners, like riveting and bolting. Fatigue damage remains one of the primary causes of failure in adhesively bonded joints. In this work, probabilistic fatigue life prediction and probabilistic damage prognostics frameworks are developed for adhesively bonded joints. The probabilistic fatigue life prediction framework calibrates the fatigue life model by quantifying uncertainty in the fatigue damage evolution relation using experimental fatigue life data. Probabilistic assessment of fatigue life is simulated through damage evolution along the bondline and Bayesian inference via the Markov chain Monte Carlo (MCMC) sampling method for inverse uncertainty quantification (UQ) in fatigue model parameters. To expedite the fatigue life simulation, a hybrid model composed of physics-based fatigue damage evolution relation and a data-driven artificial neural networks (ANNs) model is employed. The degradation of the adhesive is evaluated by the fatigue damage evolution relation which is then mapped to the strain redistribution along the bondline using the ANNs model. Once the mapping is learned by the ANNs, through data from FEA simulations, the probabilistic fatigue life prediction framework involves three successive modules: (I) fatigue damage growth (FDG) simulator, (II) inverse UQ in fatigue model parameters, and (III) confidence bounds for fatigue life prediction. The developed FDG simulator can be used for simulating fatigue degradation rapidly for a given geometric configuration under any arbitrary fatigue loading spectra. The quantified model parameter uncertainties from the framework correspond to the intrinsic statistical material properties that can be used for probabilistic fatigue life prediction in any joint configuration with the same adhesive material. The probabilistic framework is verified u

摘要： Next-generation RNA sequencing technology has revolutionized translational research. RNA-Seq enables researchers to accurately quantify and compare the gene expression levels within and across diverse biological samples. Today, petabytes of multidimensional, complex RNA-Seq data are available publicly. Researchers analyze this multidimensional data using dimensionality reduction algorithms and heat maps—however, many available tools are written in the R or Python language. These tools lack an easy-to-use interactive interface and are not suitable for large data sets. Interactivity is a key requirement for efficient data analysis of complex multidimensional data sets. Based on this, Singh et al. developed a Java tool, MetaOmGraph [Singh et al. (2020)], for fully interactive analysis of large omics data sets. This thesis extends the functionality of MetaOmGraph (MOG) software and adds dimensionality reduction features for easy and efficient analysis of bulk and single-cell RNA-Seq data. I have added Principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE) dimensionality reduction algorithms, and heat maps to MOG to facilitate multidimensional data analysis. Researchers can interactively visualize PCA, t-SNE, and heat maps by changing the final appearance of plots using various options provided in each of these. Sample filtering in MOG is now made easy and efficient for the user by creating sample subset lists. Overall, these contributions to MOG make it more user-friendly. MOG software is available at https://***/MetNet_*** and the code is at https://***/urmi-21/MetaOmGraph.

关键词： Computer science   Artificial intelligence  

摘要： Crowdsourcing is a new decentralized problem-solving paradigm for tasks that are difficult for computers but easy for humans. In the crowdsourcing framework, a data collector assigns tasks to the workers and requests their answers to the tasks. The collected data is then processed for some data analytic tasks. Although the crowdsourcing systems provide cost-effective solutions to data collection and sophisticated data analytics, it raises several trustworthy issues, which are listed below: Security of task assignment. As the tasks may contain sensitive information, it raises the security issue of how to assign these sensitive tasks to the workers, especially when the adversarial workers can collude with each other. To address this issue, I consider adversarial colluding adversarial workers who intend to extract sensitive data by exchanging information. I design a 3-step sensitive task assignment method, and perform an extensive set of experiments on real-world and synthetic datasets. The results demonstrate the accuracy and efficiency of my method. Privacy in crowd-based data collection and analysis. In some settings such as a social network analysis, a worker's local data contains not only his private information but also his neighbors'. The challenging issue is how to protect those private information. To address this challenge, I formulate a novel privacy notion named decentralized differential privacy (DDP).Then I design a new multi-phase framework under DDP that enables to perform accurate social network analysis, e.g., triangle counts with DDP guarantee. Extensive experiments using real data demonstrate that the proposed scheme leads to accurate estimates of global subgraph counts with DDP guarantee. As the result of the privacy protection, the data collected from the workers are perturbed. Hence a natural question is how to infer accurate answers from the noisy data. In this thesis, I adapt local differential privacy (LDP), a de facto privacy notion, to crowd