教学课程资源库 更多>>

关键词： Computer science  

摘要： In recent years, there has been a significant growth in the number of new malware specimens. This resulted in novel Malware Classifiers to help identify them. Many of these Malware Classifiers claim that they use some form of Machine Learning techniques to identify malware. Our task is to evaluate such claims of MLMC's as to what extent they are true and find out if MLMC's are good at identifying new malware that has never been seen before. We have explored the idea of including diversity into the malware specimen which are generated or compiled from the source code collected from different resources like theZoo and the malsource dataset. Different transformations are performed on the source code. Experiments were done with 1-2 compiled malware specimen, applying transformations on source code level and the VirusTotal's response to these transformations.

关键词： Computer science  

摘要： Machine Learning Assisted Triaging System is a web-based application. Web-based software is a popular choice for implementing today’s applications. It provides many benefits such as centralized data access, reduced costs in hardware and many others. The goal of Machine Learning Assisted Triaging System, or MLATS, is to provide recommendations to the end users as to what questionnaire versions should be used based for a given condition. MLATS communicates with a web service that hosts a trained model which acts as a block box that contains all the logic necessary to make accurate predictions. The model uses Convolutional Neural Networks to predict on a given input. A large corpus is necessary in order to properly train the model. The training data is kept in the file system directory which is processed by the deep learning tool. The tool is commonly used for multi-label text classification problems.

关键词： Computer science  

摘要： As the science of synthetic biology matures, engineers have begun to deliver real-world applications which are the beginning of what could radically transform our lives. Recent progress indicates synthetic biology will produce transformative breakthroughs. Examples include: 1) synthesizing chemicals for medicines which are expensive and difficult to produce; 2) producing protein alternatives; 3) altering genomes to combat deadly diseases; 4) killing antibiotic-resistant pathogens; and 5) speeding up vaccine production. Although synthetic biology promises great benefits, many stakeholders have expressed concerns over safety and security risks from creating biological behavior never seen before in nature. As with any emerging technology, there is the risk of malicious use known as the dual-use problem. The technology is becoming democratized and de-skilled, and people in do-it-yourself communities can tinker with genetic code, similar to how programming has become prevalent through the ease of using macros in spreadsheets. While easy to program, it may be non-trivial to validate novel biological behavior. Nevertheless, we must be able to certify synthetically engineered organisms behave as expected, and be confident they will not harm natural life or the environment. Synthetic biology is an interdisciplinary engineering domain, and interdisciplinary problems require interdisciplinary solutions. Using an interdisciplinary approach, this dissertation lays foundations for verifying, validating, and certifying safety and security of synthetic biology applications through traditional software engineering concepts about safety, security, and reliability of systems. These techniques can help stakeholders navigate what is currently a confusing regulatory process. The contributions of this dissertation are: 1) creation of domain-specific patterns to help synthetic biologists develop assurance cases using evidence and arguments to validate safety and security of designs; 2) app

关键词： Ecosystems   Agriculture   Ontologies   Sensors   Cloud computing   Computer science   Cooperative smart farming   cyber-physical systems   artificial intelligence   precision agriculture   ontology  

摘要： Cyber-Physical Systems (CPS) and Internet of Thing (IoT) generate large amounts of data spurring the rise of Artificial Intelligence (AI) based smart applications. Driven by rapid advancements in technologies that support smart devices, agriculture and farming sector is shifting towards IoT connected ecosystem to balance the increase in demand for food supply. As the number of smart farms reach critical mass, it is now possible to include AI assisted systems at a cooperative (co-op) farming level. Today, in the United States alone there are about 1,871 co-ops serving 1,890,057 member farmers. Hence, such advanced technologies and infrastructure when incorporated in the co-op farming ecosystem can immensely benefit small member farmers who operate and maintain these independent co-op entities. In this paper, we develop a connected cooperative ecosystem which defines sensors and their communication among different entities along with cloud supported co-op hub. We develop member farm and co-op ontologies to capture data and various interactions that happen between shared resources, member farms, and the co-op that are stored in the cloud. These can then help generate AI supported insights for farmers and the cooperative. Several co-op farming use case scenarios have been discussed to demonstrate the functioning of our smart cooperative ecosystem. Finally, the paper describes various AI applications that can be deployed at the co-op level to aid member farmers.

关键词： Business administration   Computer science  

摘要： In recent years, artificial intelligence (AI), especially machine learning (ML) and deep learning (DL), has represented one of the most exciting advances in science. The performance of ML-based AI in many areas, such as computer vision, voice recognition, and natural language processing has improved dramatically, offering unprecedented opportunities for application in a variety of different domains. In the critical domain of healthcare, great potential exists for a broader application of ML to improve quality and efficiency. At the same time, there are substantial challenges in the development and implementation of AI in healthcare. This dissertation aims to study the application of state-of-the-art AI technologies in healthcare, ranging from original method development to model interpretation and real-world implementation. First, a novel DL-based method is developed to efficiently analyze the rich and complex electronic health record data. This DL-based approach shows promise in facilitating the analysis of real-world data and can complement clinical knowledge by revealing deeper insights. Both knowledge discovery and performance of predictive models are demonstrably boosted by this method. Second, a recurrent neural network (named LSTM-DL) is developed and shown to outperform all existing methods in addressing an important real-world question, patient cost prediction. A series of novel analyses is used to derive a deeper understanding of deep learning’s advantages. The LSTM-DL model consistently outperforms other models with nearly the same level of advantages across different subgroups. Interestingly, the advantage of the LSTM-DL is significantly driven by the amount of fluctuation in the sequential data. By opening the “black box,” the parameters learned during the training period are examined, and is it demonstrated that LSTM-DL’s ability to react to high fluctuation is gained during the training rather than inherited from its special architecture. LSTM-DL can

关键词： Psychology   Cognitive psychology   Computer science  

摘要： Eye movements may reflect where an individual is placing their attention, but they can also reflect search conditions, search difficulty, and many other elements relating to an individual’s cognition. As individuals search for objects of interest, they are attracted to features of stimuli that overlap with their search target. These mental representations of search targets, known as target templates, allow for efficient visual search. In turn, eye movements themselves provide information concerning what an individual is searching for. Machine learning can be leveraged to utilize this information to predict information concerning an individual’s target template. The present research consists of three studies that were conducted to evaluate the viability of predicting user search targets through a machine learning approach. Experiment 1 showed that predicting a user’s search target category was possible with a limited set of abstract stimuli, while Experiment 2 demonstrated that this was possible with hundreds of real-world stimuli. Lastly, Experiment 3 supported the machine learning model constructed in Experiment 2 through a series of simulations. Taken together, the results support that machine learning can be used with eye movement data to infer an individual’s search target with high accuracy – supporting the potential use of this approach in applied settings.

关键词： Computer science  

摘要： Student retention is an important measure when in determining student success. Retention refers to the first-time full-time student from previous fall term who returned to the same university for the following fall term. Decline in retention rate have adverse effect on stakeholders, parents, and students view about the institution, revenue generated from tuition cost and obtaining outside funds. In an effort to increase retention rates, universities have started analyzing the factors that correlate with students dropping out. Many universities have identified some of these factors and are working on developing intervention programs to help students to elevate their academic performance and eventually retain them at the university. However, identifying the students who require this intervention is a very challenging task. In this thesis, we propose the use of machine learning models to identify students who are at-risk of not being retained so that university administration can successfully deploy intervention strategies at an early stage and prevent the students from dropping out. We implemented classification algorithms including feed forward neural networks, logistic regression, and support vector machine to determine at-risk students. The data to train these models was gathered from University of Nevada Las Vegas (UNLV) enterprise data warehouse: UNLV Analytics. The models were evaluated on various metrics and the results showed that logistic regression model performed best in predicting at-risk students for first-year retention and feedforward neural networks performed best in predicting at-risk students in second-year retention at UNLV's Department of Computer Science.

关键词： Computer science  

摘要： In this work, we introduced PFault, a general methodology for analyzing the reliability of High-Performance parallel file systems. PFault emulates the failure state of each storage device in the parallel file systems based on a set of well defined fault models, and enables examining the system behavior under faults systematically. To demonstrate the effectiveness of PFault, we apply PFault to analyze two representative high-performance parallel file systems: Lustre and BeeGFS. Our analysis reveals multiple cases where the widely used parallel file systems are unable to function properly under faults, even after running the default checking and repairing utilities (i.e., LFSCK and BeeGFS-fsck). For example, as the online consistency checker of Lustre, LFSCK itself may hang or trigger the rebooting of nodes. By running LFSCK, the subsequent workloads may still hang. Moreover, we are able to find a potential security vulnerability of Lustre where reading a file after faults may retrieve the content of another irrelevant file. While running LFSCK may fix the vulnerability, it leads to another resource leak problem where a portion of Lustre’s internal namespace and the storage space become unusable. To address the resource leak problem, we design and implement a simple tool called LeakCK, which can detect the leak of internal data files based on their reachability from user files. Similar reliability issues have also been observed in BeeGFS and its checker BeeGFS-FSCK. In BeeGFS, PFault triggers less fatal problems but exposes a comparable number of faults as Lustre. BeeGFS-FSCK, the consistency checker of BeeGFS, still can’t handle most of the faults. Especially when power outage caused network failure happens, BeeGFS-FSCK itself may hang or be aborted and finally fails to reconnect BeeGFS servers. On the other hand, by studying multiple versions of Lustre and comparing their behaviors, we verify that the latest Lustre has made noticeable improvement in terms of failure

关键词： Health sciences   Computer science  

摘要： Respiratory pattern is one of the main vital signs used to measure body’s basic functions. This thesis work focused on finding a reliable, robust, and affordable means of effectively extracting the respiratory signal of a subject using a single photoplethysmography fingertip sensor and various software enhancement techniques. I created a software platform with algorithms to extract respiratory information from a finger sensor signal. The novel sensor acquired samples at a rate up to 50% higher than devices on the market, complemented by a suite of algorithms that was able to establish correlations and derive the respiratory signal from photoplethysmography. As a result, a single sensor would be able to extract more than one signal, thus curbing the need for multiple devices and making tracking of bio signals that much easier.

关键词： Computer science  

摘要： The purpose of this study was to develop an artifact that supports an automated selection of financial assets to build an investment portfolio that meets multi-objective criteria and optimized for multiperiod to minimize the investment risk while maximizing the return. The research created an artifact that automates a multi-objective portfolio optimization problem in multiperiod. The research used 10 years of six different financial asset-class data. The study considered all the information points for the research period and divide the first two-thirds of the training set information and use the next one-third of the information to test the research optimization model's effectiveness. A random sampling method is used to select 30 arbitrary starting points is used to ensure that the model is not prematurely exposed to information. Solving a challenge of multi-objective portfolio optimization in a multiperiod has been a major area of research in finance, computer science, and mathematics. It is important to examine the economic environment and assets in the portfolio to optimize in multiperiod rebalancing. In this study, a multiperiod portfolio optimization was shown by utilizing multiple objectives that characterize investors. How to balance the utility of higher returns in contrast to the risk by utilizing loss aversion, liquidity, and diversification objectives also shown in this study. Another contribution of this research was the use of big data technologies to build an online portfolio optimization pipeline artifact to achieve a high reward with a minimum risk. Multiple tests, including paired sample t-test and non-parametric Wilcoxon rank-sum tests, were undertaken over a one-tailed hypothesis to show the output of the new method gives a better result compared to the output of the traditional mean-variance portfolio at 5% level of significance.