教学课程资源库 更多>>

关键词： Accessibility   Cognitive   Computer science   Dyslexia   Education   Websites  

摘要： In a study conducted by Webaim, 98.1% of sites had a detectable accessibility issue. This poses a profound challenge to the 1 billion users across the world who have a disability. This indicates that developers either are not aware of how to make the sites accessible or aware of how critical it is to make the sites usable by all users. This problem is further compounded by the lack of available resources that can educate students and future developers in making their software accessible. To address current limitations/challenges, we have developed an all-in-one immersive learning experience known as the Accessibility Learning Labs (ALL). These modules are carefully crafted to provide students a better understanding of various accessibility topics and increase awareness. They incorporate the best of all learning methods, from case studies to hands-on activities and quizzes. In this paper, we focus specifically on the cognitive module developed under the Accessibility Learning Labs. This module strives to educate students on the importance of building accessible software for users with cognitive disabilities. We discuss the pedagogical approach used to craft the components of the cognitive module and the design rationale behind the experiential activity. We investigate how the order of the reading and experiential activity affect the students' understanding of the material. To do this, we perform a study involving 28 students in 2 computer science-related courses. Our findings include:(I) The accessibility improvements made in the lab have a positive impact on the students' performance when compared to the inaccessible version (II) When the reading material is presented after the experiential activity, students have a better understanding of the cognitive accessibility principles.

关键词： Computer science   Bioinformatics   Artificial intelligence  

摘要： Finding safe and effective drugs in the form of compounds is a long and arduous campaign that can stretch to more than a decade. This is mainly exacerbated by the exponential amount of available compounds. The effectiveness of these compounds can only be confirmed via physical testing or screening. Instead of physically screening all available compounds, virtual screening is the use of computational methods to help in carefully selecting compounds for physical screens. In this thesis we explore three virtual screening options for hit finding via four projects. When historical screening data is available for a target of interest, one can leverage virtual screening methods that model the compound-target interaction data. Given a compound budget and a prospective pool, a question of interest is the effectiveness of such methods in prioritizing hit selection in a followup screen. In such a setting, which we name one round screening, we showcase in chapters 2 and 3 that such methods can succeed in finding hits that far exceed the expectation at random, even when the prospective pool exceeds a billion compounds. An alternative is iterative batched screening, whereby the compound budget is split across multiple rounds of screening. The focus in this setting is to develop a compound selection strategy with the goal of maximizing hits and the diversity of these hits at the end of the iterations. In chapter 5, we formalize this setting and propose a cluster based selection strategy that incorporates concepts from active and reinforcement learning. Following a sequence of rigorous retrospective experiments, a final strategy is used in an iterative screen on a prospective target that mimics a realistic setting. The results of this prospective iterative screen were satisfying, finding 20.28% of the hits in just 4.25% of the prospective pool. When no screening data is available for a target of interest, we focus on a particular setting where there exists compound interaction data

关键词： Robots   Aircraft   Partial differential equations   Algorithms   Error analysis   Chemical reactors   Ordinary differential equations   Controllers   Robotics   Aerospace engineering   Computer science   Engineering   Mathematics  

摘要： Dynamics models are used to develop control algorithms for many applications, including aerospace vehicles, self-driving cars, robots, industrial plants, and more. Physics-based models are particularly common, due to their interpretability and ability to efficiently encode knowledge of the system’s behavior. An ideal model accurately represents the system’s dynamics while not introducing significant complexity into the control algorithm. However, conflict between model accuracy and complexity often arises, and thus the two must be balanced to satisfy practical requirements. While lowdimensional models provide sufficient accuracy for some control applications, there are also many where high-dimensional models are required. One important application is the control of systems best modeled by partial differential equations (PDEs), such as systems with fluid flows or fluidstructure interaction (e.g. highly maneuverable or flexible aircraft), deformable structures (e.g. soft robots), and more. In practice, infinite-dimensional PDE models are generally semi-discretized to produce high-fidelity ordinary differential equation (ODE) models that are finite-dimensional. Since the dimension of these models can range from thousands to millions, standard model-based controller design can be extremely challenging. In this thesis, we propose an approach for efficiently designing high-performing controllers based on high-dimensional models. Specifically, we develop a model predictive control (MPC) algorithm for solving constrained optimal control problems that leverages high-fidelity, but low-dimensional, reduced order approximations of the original model to satisfy practical computational requirements. In the linear setting, we combine existing ideas from tube MPC with novel approaches for controller synthesis and analysis to develop a reduced order MPC (ROMPC) scheme for solving robust, output feedback control problems, and we provide theoretical closed-loop performance guarantees

关键词： Robots   Automobiles   Global positioning systems--GPS   Unmanned aerial vehicles   Algorithms   Localization   Sensors   Vehicles   Aerospace engineering   Computer science   Engineering   Robotics   Transportation  

摘要： Accurate localization is critical for many robotic systems. Localization is especially challenging for agents in environments without access to signals from a Global Navigation Satellite System (GNSS), such as underwater or in deep space. Terrain-Relative Navigation (TRN) is one approach to solving the GNSS-denied navigation problem, which works by matching sensor measurements of the environment with a known map to find the most likely position of the vehicle in map coordinates. A TRN system using a particle filter to compare sonar range measurements to the seafloor with a bathymetric map has been developed and tested on Autonomous Underwater Vehicles (AUVs) in Monterey Bay. This system has been shown to be highly effective at localizing vehicles over informative terrain, but fails to converge for vehicles over flat terrain. This thesis attempts to solve the problem of localization in flat terrain by fusing measurements between a team of vehicles acting cooperatively. Assuming vehicles can measure the relative positions of other agents and communicate in real time, vehicles over informative terrain can act as ad-hoc navigational aids to localize vehicles in flat terrain. However, the well-known problem of measurement correlation in cooperative navigation networks causes existing state estimation filters to become inconsistent and overconverge. No prior approach is able to meet the requirements of cooperative underwater navigation, specifically, a distributed algorithm with low-bandwidth communication that allows for multi-modal position estimates while remaining consistent. This thesis introduces a new filter architecture that meets these requirements for underwater cooperative TRN. By splitting the estimator into a bank of filters, with each filter uniquely assigned to a different information source, the architecture entirely eliminates one type of measurement correlation. The remaining correlation is handled using Covariance Intersection, a conservative technique

关键词： Computer science   Artificial intelligence   Robotics   Computer Engineering  

摘要： Deep learning and reinforcement learning have recently had a transformative impact on the fields of computer vision, NLP, and robotics. However, most of the recent progress have been in narrowly defined tasks where training and deployment happen under the same (or similar) conditions. This has resulted in brittle models that catastrophically fail when presented with conditions even moderately different from what they were trained on. Furthermore, current approaches are known to be data hungry, and thus may require prohibitively large datasets for many applications. How can we create intelligent agents that are data-efficient, robust, capable of broad generalization, and fast adaptation? In this thesis, we outline the importance of domain transfer as a key component to achieve the aforementioned capabilities. Domain transfer refers to the ability of an AI agent to draw upon experiences from related tasks and transfer inductive biases, enabling more efficient and proficient learning in downstream tasks. This thesis presents abstractions and algorithms to enable such domain transfer. In particular, we focus on domain transfer that arises in the context of simulation to reality transfer in robotics, learning from static offline datasets in reinforcement learning, and efficient adaptation to new tasks in the case of meta-learning. The algorithms we present enjoy rigorous theoretical guarantees and also demonstrate strong empirical results in a variety of benchmark tasks and real-world case studies spanning perception, control, and robotics.

关键词： Bioinformatics   Computer science   Biostatistics   Genetics   Computational chemistry  

摘要： Most genome-wide and transcriptome-wide association studies (GWAS, TWAS) focus on European populations; however, these results cannot always be accurately applied to non-European populations due to differences in genetic architecture. Using summary statistics from GWAS in the Population Architecture using Genomics and Epidemiology (PAGE) study, which comprises ~50,000 Hispanic/Latinos, African Americans, Asians, Native Hawaiians, and Native Americans, we perform transcriptome-wide association studies to determine gene-trait associations. Initially, we compared results using two transcriptome prediction models derived from the Multi-Ethnic Study of Atherosclerosis (MESA) populations: the African American (AFA) model and the Hispanic/Latino (HIS) model. We identified 141 unique genome-wide significant trait-associated genes. 11 of the 141 gene-trait pairs were found to have colocalized eQTL and GWAS signals and replicated in PhenomeXcan. Overall, we identified more significant, colocalized, replicated gene-trait pairs in the HIS MESA model than the AFA model. Since the largest population in PAGE is of Hispanic/Latino ancestries, TWAS with more population-matched transcriptome models, i.e. HIS rather than AFA, have more power for discovery and gene-trait replication. Following this analysis, we then compared results using three larger transcriptome prediction models derived from Multi-Ethnic Study of Atherosclerosis (MESA) populations: the African American and Hispanic/Latino (AFHI) model, the European (EUR) model, and the African American, Hispanic/Latino, and European (ALL) model. We identified 240 unique genome-wide significant trait-associated genes. We found more significant, colocalized genes that replicate in larger cohorts when applying the AFHI model to the PAGE summary statistics than the EUR or ALL model. We also found more significant gene-trait pairs using the AFHI, which identified 152 trait-associated genes, when compared to the number of associations ma

关键词： Computer science   Design   Museum studies   Cultural anthropology   Cultural resources management   Software   Native peoples   Museums   Education   Cultural heritage  

摘要： Extended Reality (XR) technologies continue to be developed by GLAM (galleries, libraries, archives, and museums) and other institutions to create cultural heritage exhibitions to preserve and transmit ancestral knowledge of a particular community. This thesis proposes a shift in the paradigm of exhibition design; one that is not only user-centered but also measures the success of an exhibit by the equity it provides the community of the cultural heritage. The equitable design framework is introduced and adapted particularly for XR development, to aid in the embedding of equitable methods within the typical methodology of software developers who may work within the cultural heritage field. Building on prior Indigenous guidelines and methodologies, the equitable design framework consists of waypoints that can be interpreted as behavioral design patterns for XR developers to help build the skill of cultural sensitivity and reciprocity into their practice. Through an ethnographic study, this thesis contributes to the creation of an explicit model of design values derived from interviews of professionals in a variety of fields who are of Native Hawaiian descent and demonstrates how it has affected the design of a XR exhibition called Wao Kiʻi. With this proposed framework and process, this thesis contributes to the ongoing conversation of ethical design and collaborative methodologies and argues for the importance of the community's own people, ancestral knowledge, and creative expression to be meaningful factors in the development and design of XR cultural heritage exhibitions.

关键词： Computer science   Optics  

摘要： Augmented and Virtual Reality (AR/VR) are emerging as next generation personal and social computing platforms. However, for AR and VR displays to be practical and mass adopted, they need to be compact and lightweight, and should support focal accommodation like everyday prescription eyeglasses. The existing commercial displays however are bulky and do not support focus cues for virtual imagery. Furthermore, users without normal vision need to wear additional pairs of prescription correction glasses to see well-focused real imagery, which is an unreasonable and awkward requirement. In my dissertation, I focus on three key areas for augmented and virtual reality (AR/VR): 1) high-quality 2D and 3D holographic near-eye display approaches compensating for severe real-world aberrations and supporting high resolution, 2) auto-focus eyeglasses for viewing both real and virtual imagery, and 3) wearable eye tracker capable of sensing pupil position, gaze direction, and optical accommodative state. The only demonstrated compact, wide-angle eyeglasses-style near-eye display (NED) for augmented/virtual imagery uses holographic techniques. However, holographic NEDs suffer from a small eyebox and poor image quality with debilitating artifacts. To overcome these, I developed a new holographic rendering framework which directly optimizes for 2D and 3D phase-only holograms, compensating for any hardware and optical aberrations in the display, producing state-of-the-art high quality true holographic imagery. The auto-focus capability is achieved by using focus-tunable lenses with focal power dynamically adjusted to also consider the user's prescription, thereby allowing for well-focused viewing of both real and synthetic imagery. This requires knowing not only the vergence but also the accommodation state of the user's eye at any instant. To achieve this, I developed comprehensive 3D eye tracking of both eyes, by analyzing the Purkinje reflections from multiple layers in the eye's cor

关键词： Artificial intelligence   Industrial engineering   Web studies   Computer science  

摘要： Web services are software components that can be exposed on the Internet and invoked using standard protocols. The syntactic properties of interfaces of Web services are described using Web Service Description Language WSDL. Research in semantic Web services aims at developing methods to describe the semantic and pragmatic aspects of Web services in addition to the syntactic aspects. These descriptions should permit formal reasoning about the capabilities of the published services, enabling software agents to interact, composite and execute automatically with services and the composition of Web services to accomplish certain defined goals. The complex structure represents the composition of Web services is called workflow. This dissertation extends existing researches about the Web Services Composition to propose and implement an automatic Web services composition framework. Our proposed framework supports automatic Web services discovery, automatic selection and execution, automated planning and semantic Web services description. The contributions of this dissertation include 6 parts. (1) In Chapters 3, 4 and 5, we introduce the Web service composition (WSC) problem, describe the application domains, and their ontologies in our framework. In these chapters, we also identify the different issues, challenges and requirements of a complete automatic Web services composition framework. (2) In Chapter 6, we describe and implement our Web services composition framework. (3) The next chapter focuses on the formalization and implementation of Web service composition engine based on AI planning techniques. (4) In Chapter 8, we describe, formalize and implement the WSC problems related to workflow selection, refinement, manipulation and re-planning engine. Specifically, we provide different approaches to these problems which satisfy different optimization requirements. (5) The next chapter deals with failure execution and workflow repair process. (6) In the last chapter, the

关键词： Computer science   Biomedical engineering   Neurosciences   Oncology   Artificial intelligence   Software   Brain cancer   Neural networks  

摘要： Even in a developed country like America the ratio of the number of hematologist and oncologist together to the number of people per doctor is 1:20,366, then imagine how much it would be in the entire world? Medical industry has progressed tremendously over the past years and, specifically, the visual and image recognition is being used for many purposes, in many fields very actively. The general Artificial Intelligence (AI) topics, such as, Neural Networks (NN) and related concepts have also gained a lot of popularity lately. For this research, we have proposed a convolutional neural networks (CNN) architecture from scratch with data augmentation, image processing approaches, and neural network pattern recognition. We have compared the pre-existing architecture VGG-16 to our CNN model to determine if the CNN model was more suitable for these types of detection problems. Although the data that we have used to train the model in our research is comparably small to any industrial application database, the proposed model displayed better accuracy and results than the VGG-16 for this kind of detection problem. Moreover, the proposed model uses less computational and memory power than the VGG-16 model. The secondary purpose of this research is to reduce the use of datasets that are from unrecognized sources. In this paper, we show how we can take the consent of the user and store the data to build a true data set for future educational and medical researchers and to retrain the models for better results.