教学课程资源库 更多>>

关键词： Search engines   Task analysis   Metadata   Mathematical model   Internet   Computer science   Tools   Bidirectional LSTM   hierarchical LSTM   document understanding   specialty search engine  

摘要： This paper introduces an automatic approach to understand the purposes of each sentence in the abstract of an academic document. Specifically, computers can label each sentence in the abstract as being related to one or several of six aspects - "BACKGROUND", "OBJECTIVES", "METHODS", "RESULTS", "CONCLUSIONS", and "OTHERS". Experimental results obtained on a real dataset show that the labeling methodology outperforms baseline methods. We also build a prototype academic search engine to demonstrate the use of this new design. Users may search for articles containing keywords related to any of these six aspects to better meet their search goals.

关键词： Computer engineering   Computer science   Electrical engineering   Artificial intelligence   Information technology  

摘要： As applications become highly data-centric, there are major challenges associated with (i) the limitations of CMOS technology for the design of on-chip (fast access, dense) memories, (ii) the practicality of transferring high volumes of data between the processor and main memory, and (iii) the security and privacy of client’s data stored in external servers, which have become a very popular demand in the last years. In this dissertation, I adopt a bottom-up perspective -i.e., from devices to architectures - to look at potential alternatives to address these issues associated with a high volume of data handling and processing. First, despite advancements enabled by CMOS scaling, as transistors shrink, quantum effects and current leakage in conventional CMOS technology becomes very significant, which results in prohibitively high static power consumption. Furthermore, low density and high leakage power associated with 2-D Static SRAMs built with CMOS make it challenging to satisfy the growing on-chip memory demands from data-centric applications. In this regard, in this dissertation, I explore emerging technologies, with a focus on Ferroelectric Field-Effect Transistors (FeFETs). Emerg- ing technologies enable the design of better on-chip memory alternatives which can improve density and/or energy consumption. Second, the high volume of data movement may result in a slowdown in the performance of data-intensive programs, and high energy consumption due to the various memory accesses needed by the application. In this work, computing-in-memory architectures (CiM) are proposed as solutions for overcoming the latency and energy problems associated with data transfer, as they enable a subset of logic, arithmetic, and memory operations associated with a given task are performed in memory (without transfers from/to a processor). Finally, like speed and energy consumption, the security and privacy of computer systems have become a critical research topic. On one hand, hardwa

关键词： Computer science  

摘要： Graphs, which encode pairwise relations between entities, are a kind of universal data structure for many real-world data, including social networks, transportation networks, and chemical molecules. Many important applications on these data can be treated as computational tasks on graphs. For example, friend recommendation in social networks can be regarded as a link prediction task and predicting properties of chemical compounds can be treated as a graph classification task. An essential step to facilitate these tasks is to learn vector representations either for nodes or the entire graphs. Given its great success of representation learning in images and text, deep learning offers great promise for graphs. However, compared to images and text, deep learning on graphs faces immense challenges. Graphs are irregular where nodes are unordered and each of them can have a distinct number of neighbors. Thus, traditional deep learning models cannot be directly applied to graphs, which calls for dedicated efforts for designing novel deep graph models. To help meet this pressing demand, we developed and investigated novel GNN algorithms to generalize deep learning techniques to graph-structured data. Two key operations in GNNs are the graph filtering operation, which aims to refine node representations; and the graph pooling operation, which aims to summarize node representations to obtain a graph representation. In this thesis, we provide deep understandings or develop novel algorithms for these two operations from new perspectives. For graph filtering operations, we propose a unified framework from the perspective of graph signal denoising, which demonstrates that most existing graph filtering operations are conducting feature smoothing. Then, we further investigate what information typical graph filtering operations can capture and how they can be understood beyond feature smoothing. For graph pooling operations, we study the procedure of pooling from the perspective of g

关键词： Energy   Applied Mathematics   Computer science   Mathematics  

摘要： Thermal properties of the underground formation govern the down hole thermal processes such as geothermal energy utilization and thermal enhanced oil recovery. In order to design and monitor the subsurface thermal processes, accurate values of the thermal properties are needed. Unless perfect descriptions of the formation and fluids are given, the results of measuring the thermal properties of formation and fluid-formation systems are of limited value. However, calculation of such properties is effortful, costly and time ***, a need exists to propose models of predicting thermal properties from other more effortlessly measurable properties. With technology development, in-situ measurements have gotten more attention to reduce the cost and time needed to measure the properties with higher accuracy. Close correlations between well bore fluid temperature and its surrounding formation have made the temperature data as an ideal source to study the underneath formation. This study firstly presents an analytical temperature model to compute wellbore fluidtemperature profile in a producing wellbore. This analytical model is then compared to anumerical model to assess the accuracy and limitation of the approach. Next, sets of numerical models are used through an inversion process developed in MATLAB softwareto predict the formation thermal properties including geothermal gradient and thermalconductivity factors as well as days of production. Finally, the approach is applied to real data from a water producing wellbore. Thetemperature profile in this wellbore was recorded by Distributed Temperature Sensing(DTS) in a producing fractured wellbore under three different flow conditions representingthree different flow rate profiles along the wellbore. The results showed that the predicted thermal properties are in good conformity with theresults obtained by the performed laboratory experiments on the same geological site byother works with around 10 % of error for

关键词： Engineering   Computer science  

摘要： Nowadays, video game players can experience many high-tech experiences such as 3D vision, light tracking, Dolby Audio, and motion capture. However, some senses are still underdeveloped, for example, olfaction. Researchers studied techniques to build an odor-release device, few researchers paid attention to how players experience the olfaction from these devices. During gameplay, sensory feedback to players is essential for constructing a player experience, and, thus, it is important to study how olfactory feedback influences player experience in video games. To study olfactory feedback, I built an odor-release device and constructed a video game with odor experience called Flappy Lemon, which is a derivative of the popular Flappy Bird game. I then tested the influence of olfactory feedback by creating two different versions, a scented version, and a regular version. In the scented version, players play Flappy Lemon with olfactory output and in the regular version without. We then handed out a post-game survey and collected in-game data to analyze the olfaction’s influence on the player experience. As the result shows, the olfactory feedback could generally improve the player’s experience. In addition, the odor-release device largely increases the game challenge - as a part of the player experience. Moreover, some in-game data shows that the odor-release device could hurt a player's performance. The positive player experience as well as the unexpected results warrant future research into the inclusion of olfactory feedback into games.

关键词： Artificial intelligence   Computer science   Astronomy  

摘要： The discovery of new exoplanets, planets outside of our solar system, is essential for increasing our understanding of the universe. Exoplanets capable of harboring life are particularly of interest. Over 600 GB of data was collected by the Kepler Space Telescope, and about 30 GB is being collected each day by the Transiting Exoplanet Survey Satellite since its launch in 2018. Traditional methods of experts examining this data manually are no longer tractable; automation is necessary to accomplish the task of vetting all of this data to identify planet candidates from astrophysical false positives. Previous state-of-the-art models, Astronet and Exonet, use deep convolutional neural networks (CNNs) with over 8.8 million parameters. In this paper, I experiment with the application of recurrent networks, attentional models, and scaling down Astronet. I have developed a CNN model with 8x fewer trainable parameters than Astronet with the same accuracy and improved precision. I also provide a CNN-LSTM model with 59x fewer parameters just 1\% behind Astronet in accuracy that, with further tuning, may also be a competitive model for particularly resource-constrained uses. All code for this research is available on GitHub.

关键词： Algorithms   User behavior   Sensors   Roads & highways   Automobile safety   Vehicles   Automotive engineering   Computer science   Engineering   Transportation  

摘要： Traditional accident-based approach is not suitable for microscopic road safety evaluation due to small sample records. Instead, Traffic conflicts are often used as surrogate measures. Conventional application of traffic conflict indicators, e.g. time to collision (TTC), are often conducted at fixed time intervals, utilising the most critical moment of an encounter. Such approach cannot translate well into location-based evaluation, which is more meaningful for drivers and policy-makers. This research proposes space-based TTC (STTC), measured at fixed space intervals, capable of identifying risks microscopically at specific spots on the roadway. Thorough comparison with TTC suggests advantages of STTC in terms of formulation, aggregation perspective and interpretation for location-based safety evaluation. For threshold calibration, an optimization algorithm addressing different severity classes of conflicts and crashes is developed. Given the calibrated thresholds and weights, cross-correlation tests are conducted, which validates STTC and more importantly discovers a dislocation effect between conflict and crash locations.

关键词： Computer science   Artificial intelligence   Education  

关键词： Computer science   Artificial intelligence  

摘要： Relationship-based access control (ReBAC) can express access control policies in terms of chains of relationships between entities, as well as attributes of entities. This increases expressiveness and flexibility that promote security and information sharing. ReBAC can be formulated as an object-oriented extension of attribute-based access control (ABAC) in which relationships are expressed using fields that refer to other objects, and path expressions are used to follow chains of relationships between objects. ReBAC policy mining algorithms have the potential to significantly reduce the cost of migration from legacy access control systems to ReBAC, by partially automating the development of a ReBAC policy from an existing access control policy and attribute data. This thesis considers problems of mining ReBAC policies from two different sources of information: a set of authorizations, typically obtained from access control lists, and attribute data represented as an object model. We introduce the first algorithm for each of these problems. First, we present four algorithms for mining ReBAC policies from a complete set of authorizations and complete information about attribute data: a greedy algorithm guided by heuristics, a grammar-based evolutionary algorithm, a simplified evolutionary algorithm that also utilizes a neural network to reduce the search space, and an algorithm based on decision trees. Second, we present three algorithms that are extensions of those mentioned, serving the purpose of mining ReBAC policies from inexact authorization (typically obtained from operation logs), and complete attribute data, including an extension of the greedy algorithm, and extensions of the two grammar-based evolutionary algorithms. Third, we present extensions of the decision tree based algorithm for mining ABAC and ReBAC policies from a complete set of authorizations and incomplete attribute data, where the values of some attributes of some entities are unknown. This is

关键词： Computer science   Decomposition   Semantic web   Clustering   Graphs   Computer security   Design   Algorithms   Knowledge representation   Software engineering   Semantics   Energy   Engineering   Web studies  

摘要： Information Networks are collections of data structures that are used to model interactions in social and living phenomena. They can be either homogeneous or heterogeneous and static or dynamic depending upon the type and nature of relations between the network entities. Static, homogeneous and heterogenous networks have been widely studied in data mining but recently, there has been renewed interest in dynamic heterogeneous information networks (DHIN) analysis because the rich temporal, structural and semantic information is hidden in this kind of network. The heterogeneity and dynamicity of the real-time networks offer plenty of prospects as well as a lot of challenges for data mining. There has been substantial research undertaken on the exploration of entities and their link identification in heterogeneous networks. However, the work on the formal construction and change mining of heterogeneous information networks is still infant due to its complex structure and rich semantics. Researchers have used clusters-based methods and frequent pattern-mining techniques in the past for change discovery in dynamic heterogeneous networks. These methods only work on small datasets, only provide the structural change discovery and fail to consider the quick and parallel process on big data. The problem with these methods is also that cluster-based approaches provide the structural changes while the pattern-mining provide semantic characteristics of changes in a dynamic network. Another interesting but challenging problem that has not been considered by past studies is to extract knowledge from these semantically richer networks based on the user-specific constraint. This study aims to develop a new change mining system ChaMining to investigate dynamic heterogeneous network data, using knowledge engineering with semantic web technologies and data mining to overcome the problems of previous techniques, this system and approach are important in academia as well as real-life app