教学课程资源库 更多>>

关键词： Applied mathematics   Information science   Computer science  

摘要： Across the globe, the frequency and size of wildfire events are increasing. Research focused on minimizing wildfire is critically needed to mitigate impending humanitarian and environmental crises. Real-time wildfire response is dependent on timely and accurate prediction of dynamic wildfire fronts. Current models used to inform decisions made by the U.S. Forest Service, such as Farsite, FlamMap and Behave do not incorporate modern remotely sensed wildfire records and are typically deterministic, making uncertainty calculations difficult. In this research, we tested two methods that combine artificial intelligence with remote sensing data. First, a stochastic cellular automata that learns algebraic expressions was fit to the spread of synthetic wildfire through symbolic regression. The validity of the genetic program was tested against synthetic spreading behavior driven by a balanced logistic model. We also tested a deep learning approach to wildfire fire perimeter prediction. Trained on a time-series of geolocated fire perimeters, atmospheric conditions, and satellite images, a deep convolutional neural network forecasts the evolution of the fire front in 24-hour intervals. The approach yielded several relevant high-level abstractions of input data such as NDVI vegetation indexes and produced promising initial results. These novel data-driven methods leveraged abundant and accessible remote sensing data, which are largely unused in industry level wildfire modeling. This work represents a step forward in wildfire modeling through a curated aggregation of satellite image spectral layers, historic wildfire perimeter maps, LiDAR, atmospheric conditions, and two novel simulation models. The results can be used to train and validate future wildfire models, and offer viable alternatives to current benchmark physics-based models used in industry.

关键词： Computer engineering   Information technology   Information science   Computer science  

摘要： Along with the acceptance of Service-Oriented Architecture (SOA) as a promising style of software design, the role that Quality of Service (QoS) plays in the success of SOA-based software systems has become much more significant than ever before. When QoS is documented as a Service-Level Agreement (SLA), it specifies the commitment between a service provider and a client, as well as monetary penalties in case of any SLA violations. To avoid and reduce the situations that may cause SLA violations, service providers need tools to intuitively analyze if their service design provokes SLA violations and to automatically guide them preventing SLA violations. Due to the dynamic nature of service interaction during the operation of SOA-based software systems, the avoidance of SLA violations requires prompt detection of potential violations before prevention takes place at real-time. To overcome the low latency time in practice, this thesis research develops an approach of using Machine Learning techniques to not only predict SLA violations but also prevent them by means of optimization. This research discusses the algorithm and framework, along with the results of the experiments, which will help to examine its usefulness for service providers working on the construction and refinement of services.

关键词： Computer science  

摘要： Current testing education tools provide coverage deficiency feedback that either mimics industry code coverage tools or enumerates through the associated instructor tests that were absent from the student’s test suite. While useful, these types of feedback mechanisms are akin to revealing the solution and can inadvertently lead a student down a trial-and-error path, rather than using a systematic approach. In addition to an inferior learning experience, a student may become dependent on the presence of this feedback in the future. Considering these drawbacks, there exists an opportunity to develop and investigate alternative feedback mechanisms that promote positive reinforcement of testing concepts. We believe that using an inquiry-based learning approach is a better alternative (to simply providing the answers) where students can construct and reconstruct their knowledge through discovery and guided learning techniques. To facilitate this, we present Testing Tutor, a web-based assignment submission platform to support different levels of testing pedagogy via a customizable feedback engine. This dissertation is based on the experiences of using Testing Tutor at different levels of the curriculum. The results indicate that the groups using conceptual feedback produced higher-quality test suites (achieved higher average code coverage, fewer redundant tests, and higher rates of improvement) than the groups that received traditional code coverage feedback. Furthermore, students also produced higher quality test suites when the conceptual feedback was tailored to task-level for lower division student groups and self-regulating-level for upper division student groups. We plan to perform additional studies with the following objectives: 1) improve the feedback mechanisms; 2) understand the effectiveness of Testing Tutor’s feedback mechanisms at different levels of the curriculum; and 3) understand how Testing Tutor can be used as a tool for instructors to gauge learning a

关键词： Computer science  

摘要： We introduce a new software framework for the development of immersive virtual reality applications. Our aim is to integrate varying developing environments into a single unified framework so that the researchers can focus on their work independent of the platform. In this work, we explore three possible use-cases utilizing our framework : Immersive volumetric visualization, virtual reality scene, and data visualization. We concentrate mostly on volume rendering because volume rendering is significantly a resource demanding task. Due to its such property, we believe it is able to suggest us the feasibility of applying the framework into various types of applications. Next, we investigate the performance of the framework when executing those applications. We apply different techniques used in volume rendering to fully examine the framework and compare them. Our results indicate that the framework can be customized to suit the purpose of 3D volumetric visualization task, as well as for virtual reality and data visualization. Lastly, we review the current limitation of the framework and propose detailed future development plans of the framework.

关键词： Health sciences   Medical imaging   Computer science  

摘要： Accurate prediction of a cancer patient’s outcome is imperative to treatment selection and decision support. Recently, automated data analysis pipelines have been explored to model and predict complex patient outcome relationships using information extracted from existing data sources; for example, radiomics features that are designed to quantify patient images. However, consideration of biases imposed by human knowledge and interactions with these systems, and their resulting models, is required to obtain robust and informative results. In this thesis, automated data analysis pipelines are developed for exploration and prevention of unintended biases. A comprehensive radiomic pipeline was established and validated during an in-depth analysis of a highly cited radiomic signature; concluding that the signature was a surrogate for tumour volume and that intensity and texture values were not pertinent for prognostication. Safeguards were subsequently suggested to refine methodologies and ensure productive radiomic developments. Based upon the recommended safeguards, automated image quality classification methods were developed. In particular, convolutional neural networks (CNN) were trained to automatically classify head and neck (H&N) computed tomography images based upon the presence of dental artifacts. The CNNs were additionally validated using six external datasets and demonstrated increased classification performance with transfer learning. The suggested safeguards were also enforced in the development of operator-controlled data analysis pipelines with different levels of human interaction requirements. The pipelines predicted loco-regional failure in H&N patients using combinations of clinical, imaging (i.e. radiomic) and radiation therapy (i.e. RTx-omic) information. These pipelines provided insights into the importance of data curation and quality, as well as operator, data and methodology bias awareness as it pertains to result interpretation in operator-con

关键词： Computer science  

关键词： Behavioral sciences   Computer science  

摘要： Active safety technologies and driving have become increasingly important and being used in our modern life. They can efficiently help drivers prevent and minimize the effects of accidents by interpreting sensor information to perceive the vehicle’s surrounding environment. Sensor information is first processed by vehicles’ central embedded system (also called active safety systems) and then used to determine the level of intervention for safe driving. However, most research is focused on either finding optimal methods for fusing and interpreting sensor information without consideration of drivers’ continuous attention and behavior, or with improving the configurations of active safety systems with high-cost components such as LIDAR, night-vision cameras, or radar sensor arrays. We find that the effectiveness of active safety technologies and systems is greatly affected by drivers’ attention and behaviors. In this thesis, drivers’ attention and behaviors are characterized and jointly studied with a novel framework based on Convolutional Neural Networks. The framework is designed for general purposes with off-the-shelf dashboard cameras to assess and predict dangerous behaviors. An identification process is developed to carefully combine driver behaviors, attention, images from the vehicle with corresponding throttle and steering angle. We then identify driver behavior patterns, and construct a machine learning model for image processing and danger assessment. Finally, a model is generated to evaluate the performance of existing and proposed methods with high fidelity before deploying on vehicles in the real world. The stable performance of our framework shows the effectiveness and advantages of driver intention identification and behavior prediction for avoiding fatal collisions and eliminate the probability of potential unsafe driving behavior.

关键词： Computer science  

摘要： Recently, a well-designed and well-trained neural network can yield state-of-the-art results across many domains, including data mining, computer vision, and medical image analysis. But progress has been limited for tasks where labels are difficult or impossible to obtain. This reliance on exhaustive labeling is a critical limitation in the rapid deployment of neural networks. Besides, the current research scales poorly to a large number of unseen concepts and is passively spoon-fed with data and supervision. To overcome the above data scarcity and generalization issues, in my dissertation, I first propose two unsupervised conventional machine learning algorithms, hyperbolic stochastic coding, and multi-resemble multi-target low-rank coding, to solve the incomplete data and missing label problem. I further introduce a deep multi-domain adaptation network to leverage the power of deep learning by transferring the rich knowledge from a large-amount labeled source dataset. I also invent a novel time-sequence dynamically hierarchical network that adaptively simplifies the network to cope with the scarce data. To learn a large number of unseen concepts, lifelong machine learning enjoys many advantages, including abstracting knowledge from prior learning and using the experience to help future learning, regardless of how much data is currently available. Incorporating this capability and making it versatile, I propose deep multi-task weight consolidation to accumulate knowledge continuously and significantly reduce data requirements in a variety of domains. Inspired by the recent breakthroughs in automatically learning suitable neural network architectures (AutoML), I develop a nonexpansive AutoML framework to train an online model without the abundance of labeled data. This work automatically expands the network to increase model capability when necessary, then compresses the model to maintain the model efficiency. In my current ongoing work, I propose an alternative met

关键词： Educational psychology   Science education   Computer science  

摘要： Introductory computer programming course is one of the fundamental courses in computer science. Students enrolled in computer science courses at the college or university have been reported to lack motivation, and engagement when learning introductory programming (CS1). Traditional classrooms with lecture-based delivery of content do not meet the needs of the students that are being exposed to programming courses for the first time. Students enrolled in first year programming courses are better served with a platform that can provide them with a self-paced learning environment, quicker feedback, easier access to information and different level of learning content/assessment that can keep them motivated and engaged. Introductory programming courses (hereafter referred to as CS1 and CS2 courses) also include students from non-STEM majors who struggle at learning basic programming concepts. Studies report that CS1 courses nationally have high dropout rates, ranging from anywhere between 30-40% on an average. Some of the reasons cited by researchers for high dropout rate are lack of resource support, motivation, lack of engagement, lack of motivation, lack of practice and feedback, and confidence. Although the interest to address these issues in computing is expanding, the dropout rate for CS1/CS2 courses remains high. The software engineering industry often believes that the academic community is missing the mark in the education of computer science students. Employers recognize that students entering the workforce directly from university training often do not have the complete set of software development skills that they will need to be productive, especially in large software development companies.

关键词： Computer science  

摘要： Although not strictly necessary in all color image processing applications today, color quantization still plays an important role in certain, typically hardware constrained, applications. In this thesis, a novel color quantization method based on MacQueen’s k-means algorithm, is proposed and compared to the more popular batch k-means algorithm. The proposed method uses the maximin initialization method and quasi- random sampling to achieve high quality, fast, and deterministic results. In comparison to other well-known color quantization methods, the proposed method achieves very competitive results while being much faster.