教学课程资源库 更多>>

关键词： Computer science  

摘要： Social indicators are numerical measures that describe the well-being of individuals or communities; social indicators span multiple dimensions including food and water, safety, labor, infrastructure, transportation, etc. Taken as a time series, it is possible to use social indicator data for several predictive tasks including predicting social unrest, flow of refugees, and even armed conflict. Thus far, such indicators have been generated on a periodic basis by organizations such as the United Nations and The World Bank based on tedious and expensive data collection methodology. Due to the availability of massive amounts of open source data including social media, satellite data, commodity prices, traffic data, it should now be possible to (i) map data sources to indicators, (ii) generate social indicators in (near) real-time across the global, and (iii) develop models that capture the interaction of indicators for specific predictive tasks. It is critical to develop models that take as input these heterogeneous data sources and derive representations automatically, without relying heavily on heuristic feature engineering. The modeling task is challenging since: (i) event like social unrest and conflict, is due to multiple instigating factors and can rarely be attributed to a single factor and (ii) the diversity of data types (structured, unstructured, numerical, categorical), geospatial granularity, update rates, etc. Furthermore, while efforts have been made to improve the performance of predictive models in terms of accuracy, explanations are rarely provided. In this dissertation, we present a novel framework to combine heterogenous data sources to model social indicators covering multiple topics and data types, solve predictive tasks, and generate natural language descriptions which serve as the explanations. We focus on two specific tasks, namely, civil unrest prediction, and risk assessment and violence prediction. Extensive experiments and evaluation are pre

关键词： Computer science  

摘要： The increased global salience of social media has fueled the propagation of various forms of toxicity. As digital technologies and interactive websites proliferate, Online Social Networks (OSNs), once regarded as safe havens for sharing information and providing mutual support among groups of people, have now become breeding grounds for spreading toxic behaviors, political propaganda and radicalizing content. Numerous platforms are trying to combat this phenomenon by training computational methods that are capable of automatically recognizing these toxic contents and removing them from the user-generated text on their platforms. However, given the immensity and speed of content posted on online platforms, identifying and deterring these behaviors at scale remains challenging. To facilitate research in this direction, in this dissertation, I present four studies towards assessing the role of social media platforms in the propagation of toxicity. Specifically, this dissertation aims to answer the following research questions: RQ1: What is the relationship between discussion themes and the pervasiveness of online toxicity? RQ2: Can latent features of initial toxicity be identified and used to predict further toxicity? RQ3: Is toxicity contagious? RQ4: Are toxic users/themes clustered/segregated on social media? To answer these questions, in study 1, I outlined a methodology for identifying and scoring toxicity within the user-generated content posted on YouTube. This analysis was able to shed light on how the existence and magnitude of toxicity changes when there are shifts in the narrative. In study 2, I leveraged the Non-negative Matrix Factorization (NMF) technique to describe the relationships between the impacts of toxicity of a video on its comments. This novel application of the NMF technique allowed toxicity predictions by utilizing the latent features generated based on a prior toxicity matrix. To determine if toxicity is contagious, in study 3, I proposed a t

关键词： Genetics   Artificial intelligence   Computer science  

摘要： The usage of Quadcopter in commercial fields has evolved significantly due to its phenomenal development. However, controlling the movements of a Quadcopter is a demanding task due to its complex dynamics. The usage of the Proportional-Integral-Derivative(PID) controller for stability control is quite challenging in regards to the complexity of Quadcopter's nonlinear structure. Conventional methods like Ziegler-Nichols(ZN) for tuning the PID controller for a Quadcopter do not provide efficient performance and might also cause the system to be severely damaged. In this thesis, we are addressing the problem of the controlling a Quadcopter using Metaheuristic-based PID controller. Multi-Objective Fitness Function is proposed to reduce the overall time of the step response effectively. Path planning is one of the important concepts for a Quadcopter to move from one point to another point effectively. A novel Neighborhood Search Genetic Algorithm (NSGA) is presented for path planning by balancing the diversity inside the Genetic Algorithm using a Neighborhood Search to produce an efficient path. The performance of the NSGA has been compared to traditional $A^{*}$, standard GA, and PSO. NSGA produced superior results in terms of cost.

关键词： Computer science  

摘要： Visual tracking is one of the fundamental problems in computer vision. This topic has been a widely explored area attracting a great amount of research efforts. Over the decades, hundreds of visual tracking algorithms, or trackers in short, have been developed and a great packs of public datasets are available alongside. As the number of trackers grow, it then becomes a common problem how to evaluate who is a better tracker. Many metrics have been proposed together with tons of evaluation datasets. In my research work, we first make an application practice of tracking multiple objects in a restricted scene with very low frame rate. It has a unique challenge that the image quality is low and we cannot assume images are close together in a temporal space. We design a framework that utilize background subtraction and object detection, then we apply template matching algorithms to achieve the tracking by detection. While we are exploring the applications of tracking algorithm, we realize the problem when authors compare their proposed tracker with others, there is unavoidable subjective biases: it is non-trivial for the authors to optimize other trackers, while they can reasonably tune their own tracker to the best. Our assumption is based on that the authors will give a default setting to other trackers, hence the performances of other trackers are less biased. So we apply a leave-their-own-tracker-out strategy to weigh the performances of other dierent trackers. we derive four metrics to justify the results. Besides the biases in evaluation, the datasets we use as ground truth may not be perfect either. Because all of them are labeled by human annotators, they are prone to label errors, especially due to partial visibility and deformation. we demonstrate some human errors from existing datasets and propose smoothing technologies to detect and correct them. we use a two-step adaptive image alignment algorithm to find the canonical view of the video sequence. then use d

关键词： Computer engineering   Electrical engineering   Artificial intelligence   Computer science  

摘要： Over the past decade, machine learning (ML) with deep neural networks (DNNs) has become extremely successful in a variety of application domains including computer vision, natural language processing, and game AI. DNNs are now a primary topic of academic research among computer scientists, and a key component of commercial technologies such as web search, recommendation systems, and self-driving vehicles. However, factors such as the growing complexity of DNN models, the diminished benefits of technology scaling, and the proliferation of resource-constrained edge devices are driving a demand for higher DNN performance and energy efficiency. Consequently, neural network training and inference have begun to shift from commodity general-purpose processors (e.g., CPUs and GPUs) to custom-built hardware accelerators (e.g., FPGAs and ASICs). In line with this trend, there has been extensive research on specialized algorithms and architectures for dedicated DNN processors. Furthermore, the rapid pace of innovation in DNN algorithm space is mismatched with the time-consuming process of hardware implementation. This has generated increased interest in novel design methodologies and tools which can reduce the human effort and turn-around time of hardware design. This thesis studies how low-precision quantization and structured matrices can improve the performance and energy efficiency of DNNs running on specialized accelerators. We co-design both the DNN compression algorithms and the accelerator architectures, enabling us to evaluate the impact of our ideas on real hardware. In the process, we examine the use of high-level synthesis tools in reducing the hardware design effort. This thesis represents a cross-domain research effort at efficient deep learning. First, we propose specialized architectures for accelerating binarized neural networks on FPGA. Second, we study novel high-level synthesis techniques to reduce the manual effort in FPGA accelerator design. Third, we sho

关键词： Computer engineering   Electrical engineering   Computer science  

摘要： Since the early 1990s, Single flux quantum (SFQ) logic has been wielded as a worthy alternative to power efficient high performance computing. Many significant circuit demonstrations have been made in this technology, such as single gate clock speeds of upto 750 GHz, 8-32 bit adders, multipliers, ALU at clock speeds of 20 to 30 GHz and Analog-to-digital converters with sample rates of 40 GHz , all by several research efforts. However, this new technology has presented some design challenges as well: SFQ does not have the same density as CMOS circuits. They still need an effective power efficient solution for data storage , and most importantly do not yet have a universally accepted optimal circuit design approach. More specifically, the inherent synchronous nature of all logic gates generally used in this technology makes it tedious to obtain a direct synthesis and place and route process from Behavioral Hardware Description Language (HDL) designs. The extensive use of clock trees as a result of every gate operation being pipelined often results in jitter problems and interconnect delays on the order of 100ps. Thus, although, there have been significant recent strides in research in developing open source software to integrate an automated design flow into this technology, practical fabricated circuits have still mostly relied on intensive human design intervention to maintain operational feasibility. In light of such design challenges, a fundamentally different approach to logic gate design in SFQ has been proposed in the current work so as to make them asynchronous in nature through local self timing. This design change now makes it possible to utilize established and commercial EDA tools from vendors such as Synopsys and Cadence into following the automated design flow from synthesis to place and route in SFQ. Further in this work, a novel Auto Placement and Routing (APR) methodology is proposed for HDL based designs utilizing an asynchronous SFQ cell library. Th

关键词： Artificial intelligence   Computer science  

摘要： As the global population rises, advancements in plant diversity and crop yield is necessary for resource stability and nutritional security. In the next thirty years, the global population will pass 9 billion. Genetic advancements have become inexpensive and widely available to address this issue; however, phenotypic acquisition development has stagnated. Plant breeding programs have begun to support efforts in data mining, computer vision, and graphics to alleviate the gap from genetic advancements. This dissertation creates a bridge between computer vision research and phenotyping by designing and analyzing various deep neural networks for concrete applications while presenting new and novel approaches. The significant contributions are research advancements to the current state-of-the-art in mobile high-throughput phenotyping (HTP), which promotes more efficient plant science workflow tasks. Novel tools and utilities created for automatic code generation, maintenance, and source translation are featured. Promoted tools replace boiler-plate segments and redundant tasks. Finally, this research investigates various state-of-the-art deep neural network architectures to derive methods for object identification and enumeration. Seed kernel counting is a crucial task in the plant research workflow. This dissertation explains techniques and tools for generating data to scale training. New dataset creation methodologies are debuted and aim to replace the classical approach to labeling data. Although HTP is a general topic, this research focuses on various grains and plant-seed phenotypes. Applying deep neural networks to seed kernels for classification and object detection is a relatively new topic. This research uses a novel open-source dataset that supports future architectures for detecting kernels. State-of-the-art pre-trained regional convolutional neural networks (RCNN) perform poorly on seeds. The proposed counting architectures outperform the models above by focus

关键词： Knowledge based systems   Joining processes   Encyclopedias   Electronic publishing   Internet   Computer science   Institution name disambiguation (IND)   knowledge graph   entity linking  

摘要： The number of papers has exploded as more and more people and more types of institutions participating in scientific research. At the same time, institution name disambiguation (IND) is getting more sophisticated, which is critical for research assessment, scholar alignment, etc. Previous knowledge-based and rule-based methods require knowledge and rules prepared in advance, which cannot cope with growing and changing data and learning rules, especially for data with a long period and abundant sources. This paper proposes an automatic learning framework to solve the problem, which is based on entity linking, entity type recognition, candidate generation, and result selection. Experiments show that precision and recall is much higher than the traditional method, ELAD learns more knowledge from the knowledge graph, and it can deal with ever-changing and ever-increasing data. What & x2019;s more, it solves many problems that cannot be solved by traditional methods: the connection between institution entities, mistakes correction, and the reduction of manual and pre-prepared knowledge. At last, for the case study, we develop two applications based on ELAD which proves its reliability.

摘要： 4 | BLACK HISTORY BULLETIN VOL. 83, NO. 2 (EXTENDED DIGITAL VERSION) 83 No.2 FOREWORD BLACK FAMILY CULTURE & IDENTITY: THE POETRY & RHYTHM OF BEING By Alicia L. Moore and La Vonne I. Neal This issue of the Black History Bulletin (BHB) has three co-guest editors—Regina A. Lewis (Professor of Communication), Katherine Scott Sturdevant (Professor of History), and Stephen Collins (Professor of Communication/Rhetoric)—whose collaborations resulted in a BHB issue that lives up to their motto, “Communicate to Elevate.” The theme for this issue is The Black Family: Representation, Identity, & Diversity. Authors’ contributions include topics such as (1) exploring the metaphoric “heartbeat of the Black family” through the lens of endarkened epistemology; (2) Black family history and genealogy; (3) Black families homesteading the Great Plains; (4) the Black military family; and (5) Black entrepreneurial families. To support the action-based initiatives of our readers and online teaching initiatives, the co-guest editors included: 1. A special feature, contributed by Amery Stapleton: “Classroom Library Resource Toolkit: How to Build a Classroom Library with Books to Highlight the Culture, Strength, & Integrity of the Black Family.” 2. An online teaching guide, contributed by the co-guest editors: “Teaching History Online Using Resources from the Association for the Study of African American Life and History (ASALH).” Expressively Black Expressively Black: The Cultural Basis of Ethnic Identity is a groundbreaking book co-edited by Dr. Geneva Gay and Dr. Willie L. Baber in 1987 that continues to serve as a research-informed blueprint for understanding Black culture and identity development. We read Expressively Black as graduate students, and it is worth noting that we danced through the book chapters with ease because they appealed to our poetic and rhythmic ways of knowing. For example, we read chapters titled “The Artistry and Artifice of Black Communication,” “From Backwoods

关键词： Computer science   Apache Spark   Big data   Memory management   Memory safe   Ownership   Rust  

摘要： Planning optimized memory management is critical for Big Data analysis tools to perform faster runtime and efficient use of computation resources. Modern Big Data analysis tools use application languages that abstract their memory management so that developers do not have to pay extreme attention to memory management strategies. Many existing modern cloud-based data processing systems such as Hadoop, Spark or Flink use Java Virtual Machine (JVM) and take full advantage of features such as automated memory management in JVM including Garbage Collection (GC) which may lead to a significant overhead. Dataflow-based systems like Spark allow programmers to define complex objects in a host language like Java to manipulate and transfer tremendous amount of data. System languages like C++ or Rust seem to be a better choice to develop systems for Big Data processing because they do not relay on JVM. By using a system language, a developer has full control on the memory management. We found Rust programming language to be a good candidate due to its ability to write memory-safe and fearless concurrent codes with its concept of memory ownership and borrowing. Rust programming language includes many possible strategies to optimize memory management for Big Data processing including a selection of different variable types, use of Reference Counting, and multithreading with Atomic Reference Counting. In this thesis, we conducted an experimental study to assess how much these different memory management strategies differ regarding overall runtime performance. Our experiments focus on complex object manipulation and common Big Data processing patterns with various memory man- agement. Our experimental results indicate a significant difference among these different memory strategies regarding data processing performance.