教学课程资源库 更多>>

关键词： Computer science  

关键词： Computer Science  

摘要： As more and more devices connect with each other and exchange information through wireless interfaces, it has become increasingly important to be able to accurately and efficiently access the availability and capacity of the wireless spectrum available. The wireless channel dictates if, when, where, and how much data devices can exchange in a network. This work presents systems and studies that explore ways of making the wireless channel predictable across space and time, and generating highly accurate predictions about the channel over different frequency bands and antennas on a device. Variability in the wireless channel arises from many sources. The first source of channel variability that this work explores is that due to unpredictable interference from other devices or users in the network. The ability of the wireless channel to reliably communicate data is diminished in the presence of interference. Analyze of the interference in a network under different traffic models shows how the behaviour of individual devices affects the variability of interference over different locations in the network and over different time slots. Building upon that observation, this work presents a decentralized technique for picking transmission power and rate that devices can use to ensure more predictable levels of interference over the entire network. That in turn allows devices in the network to make reliable predictions about the interference its signal will observe at its intended receiver's location. Additionally, since the interference power is not expected to change suddenly, it can expect a more reliable link between itself and its intended receiver. Another significant source of variability in the wireless channel is the physical environment in which the signal is transmitted. The physical world intricately impacts the way a signal is reflected, refracted, or attenuated on its way to its intended receiver. If one could identify the attributes of the physical medium throu

关键词： Computer science   Cloud   DDoS   Easy-to-Adopt   Low-Cost   Network Security   Usability  

摘要： Security is always a big concern. According to the statistics, there are over 80,000 cyberattacks per day or over 30 million attacks per year. To make the Internet safe, both the industry and academia propose many solutions. However, these security solutions mainly concentrate on being effective, and ignore the other two features: deployment cost and usability. Therefore, though many works have been proposed to improve security, attacks still happen frequently. Our goal is to improve network security with low-cost and easy-to-adopt solutions. In this thesis, we choose Distributed Denial-of-Services (DDoS) attack and I/O path malware attack as two representatives. Fueled by IoT botnets and DDoS-for-Hire services, DDoS attacks have reached a record high volume, and launching such attacks is increasingly easy and cheap. We speculate the main reasons why existing solutions still leave DDoS as the top threat are 1) Commercial DDoS protection services are costly. 2) Solutions that require upgrading the core Internet architecture turned out to be extremely difficult to deploy. Similarly, modern operating systems enable user-level malware to log a user's keystrokes or scrape a user's screen output, which usually contains user sensitive data. Solutions with trusted hardware, virtual machines, and mobile phone facilitation all have high costs of deployment and usability for non-expert users. In this thesis, we present our low-cost and easy-to-adopt solutions to these two attacks. Specifically, 1) Dynashield, an on-demand DDoS defense architecture built on top of different cloud services. Dynashield introduces lower financial cost than Protection-as-a-Service product like Cloudflare, and is easier to adopt than network architecture based solutions. 2) Switchman, a framework to protect a user's I/O paths against user-level malware attacks stealing sensitive privacy data. Switchman helps non-expert users protect their sensitive data. It is easier to adopt than trusted hardware s

关键词： Artificial intelligence   Computer science  

摘要： Language models are either trained only on a repository data or post-trained using the repository after having been trained on a huge dataset like Wikipedia. Either way, since the distribution of the repository data (usually a domain-specific corpus) and the real-world distribution of concepts (such as classes in a classification application) are rarely equivalent, the accuracy of the language model is reduced. That is usually due to the Inadequacy of Knowledge (IoK) of the domain-specific corpus (referred to as the local knowledge), relative to the real-world, universal knowledge. To address this IoK issue, depending on the language modeling technique, whether traditional or recent, different methods are proposed in this dissertation to combine efficiently the local knowledge with the universal semantics and improve the performance of various text mining tasks. Considering traditional language modeling such as bag of words, two novel techniques are proposed to combine two sources of knowledge: one technique for document classification and one for document clustering. For classification, a novel feature weighting function is proposed which calculates the weight of each feature using the discriminating power of the feature derived from the local and the universal sources of knowledge. For document clustering, where no labels are available, a different technique is introduced to combine the similarities of each pair of documents, where the similarities are derived from the local and the universal knowledge. The performance of the proposed methods for document classification and document clustering is evaluated on some widely-used classification and clustering algorithms and on a number of standards datasets. The evaluation results show that the performance of both document classification and clustering techniques is significantly improved using our proposed methods. Also, we consider recent language modeling (or rather word embedding) techniques such as Word2Vec, GloV

关键词： Computer science  

摘要： Data extraction pipelines extract, transform, and integrate data from multiple sources. Such pipelines often engage multiple groups of people such as developers, annotators, data analysts, subject matter experts (SMEs), and end-users who consume the final outputs. In this dissertation, we explore techniques from research in program synthesis, database systems, and human-computer interaction to build human-in-the-loop tools that construct and debug data extraction pipelines. We will describe three tools, SEER, Texture, and WhyFlow, which target challenges across the data extraction pipeline. Constructing data extraction pipelines is particularly time-consuming and complicated for non-programmers who have to encode their domain knowledge into the pipeline’s code modules. SEER is an information extraction tool for non-programmers: it synthesizes easy-to-understand extraction rules from a handful of user-highlighted examples. Texture supports data extraction from English print documents: it provides a collaborative framework for structure identification over print document collections. Using a structured-based extraction language, end-users can construct rules that extract data from specific structures such as tables, paragraphs, lists, etc. Finally, this dissertation discusses the challenges in debugging errors identified in the outputs of a data extraction pipeline. End-users may not be familiar with all the code modules involved, which makes it difficult to pinpoint the source of the errors propagated downstream. WhyFlow synthesizes explanations over the errors in order to differentiate error flows from non-erroneous flows and zero in on the most relevant aspects of the error flows.

关键词： Games   Task analysis   Collaboration   Licenses   Learning management systems   Companies   Student activity   gamification   higher education  

摘要： Students & x2019;participation and motivation are of great importance in their learning process. The decrease in attendance and difficulties in stimulating students & x2019;activity makes it necessary to find new methodologies that can solve these problems. The use of game mechanics in non-ludic environments (Gamification) has begun to be of great interest in research, since it could increase the motivation and therefore the activity of the students. This study tries to verify if there is an existing relationship between gamification and a possible increase in student activity or between the cessation of gamification and student activity. In addition, it evaluates whether a greater student activity corresponds to a greater learning improvement. In the obtained results, no significant differences were found between the methodology or the activity with a learning improvement. According to the study, the simple use of gamified elements does not necessarily imply a solution to the problems posed. The students & x2019;activity in a course of these characteristics does not imply a better learning improvement, the importance lies in the quality of the activity generated and not in the quantity.

关键词： Computer science   Feld's friendship paradox   social computing   social network  

摘要： Feld & x2019;s friendship paradox is a widely accepted observation that states that the friends of any particular individual tend to have more friends on average than the individual has. Due to its implications for information transmission, the friendship paradox has become a generalized paradigm in many disciplines. However, how many of an individual & x2019;s friends have more friends than the individual does is still unknown, yet interesting. In this paper, we revisited the Feld & x2019;s friendship paradox and we found that only a limited number of a person & x2019;s friends have more friends than the person herself has. This conclusion was reached from empirical studies using real-world networks and is true regardless of the number of one & x2019;s neighbours, which contradicts the intuitive deduction from the friendship paradox. For one thing, if a person is unpopular, the number of her friends that have more friends than she does grows with the number of friends that she makes. This observation crystallizes the tenable margin of the friendship paradox. In another case, if a person is popular, as she acquires more friends, fewer of her friends have more friends than she does. This finding suggests an observation bias in the friendship paradox, which makes individuals feel less popular than their friends. Besides enriching our knowledge of the friendship paradox in psychological science, the findings reported here are also beneficial for technical areas such as large-scale triangle discovery. Although the friendship paradox was proposed from the perspective of the number of neighbours, the findings reported here can shed light on theories and applications from different disciplines like information, happiness, obesity, and extroversion, to name a few.

关键词： Agriculture   Artificial intelligence   Computer science  

摘要： In Corpus Christi, Texas the United States Department of Agriculture (USDA) funded Texas A&M Agrilife research on large farmlands with hundreds of individual cotton vegetation plots. Each plot is planted uniformly in rows but not all plots grow at the same rate. Every week the plots are photographed using an Unmanned Aircraft System (UAS) ying at a height of 100 feet to record and evaluate growth for various reasons. The research scientists and farmer's current method of localizing individual plots of vegetation within an image has proven to be very time consuming and inefficient. The algorithm developed in this paper automates the localization process under sunny conditions. The algorithm uses the Hue, Saturation, and Value (HSV) color space during the preprocessing stage to provide a binary image that indicates where each green pixel is located. Various OpenCV functions are then used to automate the crop localization process. Minimum and Maximum threshold values are set for ltering by size sections of the algorithm. Then, morphological operations are employed to further rene the regions of interest. The Connected Components function is used to determine how large each remaining object is and that size is then used to determine how large each localizing polygon will be drawn. After the size of each object is found, the size of each localizing polygon to be drawn is evaluated and split into smaller polygons whenever necessary. Not only the developed algorithm able to detect and classify cotton crop locations quickly but it is able to handle various complex situations. The developed method was evaluated by its accuracy, precision, and recall, which were 92.4%, 100%, and 92.4% respectively.

关键词： Information technology   Artificial intelligence   Computer science  

摘要： Previous work has shown that machine learning techniques tend to work well on homogeneous datasets, i.e., data extracted from a single source domain e.g. financial reports in a newspaper. However, the performance suffers when the training and testing datasets are derived from different domains. The problems arising from domain differences for data is referred to as domain adaptation. This thesis investigates domain adaptation methods for dependency parsing and part of speech tagging. The goal is to build a robust and generalizable system by creating domain-specific experts which aid in parsing from a heterogeneous dataset and simultaneously focus on ameliorating the problems arising from differences in annotation styles of different domains, specifically for dependency parsing. For the former problem of multi-domain adaptation, I utilize probabilistic topic models to generate domain experts. I generate topic models to automatically determine domains in the heterogeneous dataset and then train an expert for each of the domains. The test sentences are then asynchronously assigned to the appropriate expert based on similarity. This technique shows a considerable improvement in performance for both the tasks. To mitigate the problem of annotation differences in domains, I formulate a system based on transformation-based error driven learning. I use this method to learn dependency arc and label corrections in a target domain, based on errors made by the source domain parser. This method proved capable of reducing both arc and label errors considerably. This method is also flexible enough to be applied to any domain or language without any major changes to the algorithm.

关键词： Computer science  

摘要： The recent emergence of new successful two-sided online platforms has transformed the concept of a marketplace. Numerous two-sided mechanism design problems, including those studied in this thesis, are motivated by these platforms. Similar to any other mechanism design problem, we wish to optimize an objective in the presence of selfish agents. However, there are unique features to a two-sided market, such as supply uncertainty and the need for ensuring budget balance, which make these problems particularly challenging. In our problems, we design models to capture the two-sidedness, identify the challenges, and use algorithmic techniques to solve these problems. We start with a Bayesian single item auction with n independent buyers. For this problem, we show how the existence of a trusted intermediary can result in a better outcome for buyers without hurting the seller's revenue. In this model, the intermediary gets to see the true valuations of buyers and will reveal some information after that in the form of a signal. The intermediary does not have any control over agents after sending this signal, and any agent only maximizes their utility. This essentially means that the seller will run an optimal auction conditioned on receiving any signal. For this problem, we design approximately optimal ways of revealing information. The previous problem is an interesting model to show how a platform can mediate in the market and improve the outcome for every agent. That problem is a single item static problem. Next, we focus on pricing in two problems with multiple dynamic agents on the seller side. The first problem is an extension of the multiple-choice prophet inequality to the setting in which each item might disappear after an a priori unknown amount of time. This can be viewed as a way to model the supply uncertainty arising because of the fact that different sellers might depart after some time. Considering the importance of prophet inequalities in online posted pric