教学课程资源库 更多>>

关键词： Pipes   Algorithms   Alternative energy   Renewable resources   Optimization   Decision making   Natural gas   Computer science   Energy  

摘要： Nowadays, transmission system operators of energy networks (TSO) have to enable secure energy supply even under challenging transport situations and increasing demand. Hence, they seek a stable network control and regularly expand the capacity of their networks. The challenging questions arise of how to obtain safe network operations for fixed network infrastructure and how to extend the network capacity at minimum cost. In this thesis, we formulate these two questions as mathematical optimization problems and present solution approaches that scale to real-world instances of size and complexity encountered in practice. Model formulations of both problems are located in the field of Mixed-Integer Nonlinear Programming (MINLP). However, global state-of-the-art MINLP solvers are not mature enough to solve or even find primal solutions on large-scale real-world instances since the resulting degree of nonconvexity and nonlinearity pose principal difficulties. For this reason, we develop novel model formulations and optimization algorithms that significantly improve the performance of the solver SCIP. The first part of the thesis focuses on the optimization of stationary network expansions by building new pipelines in parallel to existing ones, so-called looping. Based upon a model reduction approach for multiple loops, we introduce new models for the discrete and split-pipe looping paradigm and compare them with existing models in the literature, both theoretically and empirically. It turns out that our novel split-pipe model performs best in several respects: running time, number of instances solved, and cost savings by up to 7000% over the discrete models. To further improve the performance, we analytically calculate the convex envelope of the nonconvex, nonlinear constraint function f(x,y) = y sgn(x) |x|^a, which models the physical effect of the expansion. In this way, we considerably tighten the convex relaxation of the network expansion problem. The resulting imple

关键词： Oceans   Software   Cameras   Biologists   Collaboration   Sensors   Marine biology   Internet of Things   Virtual communities   Emotions   Virtual reality   Climate change   Marine conservation   Biological oceanography   Biology   Computer science  

摘要： This thesis explores a range of Virtual Reality-related technologies in order to solve problems and expand the State-of-the-Art in the fields of Remote Collaboration, Virtual Reality, and Ocean Awareness. It also explores and harnesses the power of supporting technologies, mostly, the Internet-of-Things. The line that separates our lives from Virtuality is becoming more and more blurred. With the rapid expansion of Virtual Reality, new problems arise, both from a technical and theoretical point-of-view. This thesis explores Virtual Reality from a low-level, ”cold metal” side - such as tracking - and a high-level, ”warm feelings” side - such as emo tions. With Virtuality in mind, this thesis tries to improve Remote Collaboration based on a real-world problem of working engineers that need to cooperate in real-time en gineering endeavors while not being present in the same physical space. It explores Gamification and a mix of Augmented and Virtual Reality to improve the experience quality of remote working. Finally, this thesis takes an approach to Ocean Awareness with the help of holography, green building, and the Internet-of-Things in order to build public displays that rely on data that is collected from the marine fauna, transport it, and display it in near-real-time. This thesis also presents and discusses a set of lessons learned during these ex periments and outlines a set of design recommendations for these contexts. It has a direct impact on the industries that depend on remote collaboration as well as indus tries that need novel ways to persuade users such as ocean awareness and human well-being and improves the State-of-the-Art in Virtual Reality

关键词： Semantics   Indexes   Internet   Search problems   Search engines   Computer science   Task analysis   Academic search   knowledge graph   inverted index   structure search   citation networks analysis  

摘要： The rapid growth in the number of scholarly documents on the Web and in other digital platforms makes it challenging for researchers to find research publications most relevant to their information needs. This challenge has been mitigated to a greater extent by the major scholarly retrieval systems, such as Google Scholar, Semantic Scholar, PubMed, CiteSeerX, and others. The reason for the success of these retrieval solutions lies in the advances in ranking approaches. However, the existing studies advocate for the fact that we are still far from the method's effectiveness ceiling, leaving ample room for further improvement to meet the scholarly needs of users. The existing methods adopt different approaches;some use classical Information Retrieval (IR), others use semantics-aware methods, including Knowledge Graph (KG) to support scholarly search. However, we hypothesize that combining the best of both worlds can further improve search relevance. In this context, this work incorporates inverted index from the classical IR with BM25 as the weighting scheme, combined with Citation Networks Analysis (CNA) for the baseline search results, which are then re-ranked by passing the selected entities from the top-k initial search results as the search query to the KG. This way, not only the textual content but also the structural semantics of the research publications are well exploited in the retrieval processes. The goal is to exploit IR and KG-based retrieval techniques to gain insights into the behavior of both textual and structured information in the strategic ranking of scholarly articles. The proposed solution has been evaluated using the ACL Anthology Network (AAN) dataset. The results show that the proposed technique can comparatively improve the retrieval performance in terms of Normalized Discounted Cumulative Gain (nDCG) and precision rates.

关键词： Web studies   Computer science   Artificial intelligence  

摘要： Social media platforms, such as microblogging services and online forums, are becoming increasingly popular, profoundly revolutionizing how people share information and voice opinions. Due to the wide availability of mobile devices and easy connectivity, millions of user-generated messages are produced on a daily basis, leading us to the information explosion era. As a result, the current decade has witnessed a pressing demand for automatically digesting the large volume of social media data and discovering its crucial content. To this end, keyphrase prediction, which aims to summarize a social media post into a set of succinct keywords (or hashtags), receives growing attention in the social media research community. Previous progress made in this field has mainly focused on either extraction-based or classification-based approaches, which are limited in that they cannot predict keyphrases absent in the source text or the predefined candidate list. To overcome this limitation, in this thesis, we study neural keyphrase generation methods that enable new keyphrases to be created for social media posts. In contrast to early methods relying on handcrafted features, we take advantage of recent advances in deep learning and employ neural network-based frameworks that allow effective representation learning in a data-driven manner. More importantly, to alleviate the data sparsity issue widely exhibited in unstructured social media posts, we propose to enrich contexts via either implicitly exploiting the post-level latent topics or explicitly leveraging the user replies or the accompanying images.

关键词： Computer science   Computer engineering   Operations research  

摘要： In order to perform a large variety of tasks and achieve human-level performance in complex real-world environments, an intelligent agent must be able to learn from its dynamically changing environment. Generally speaking, agents have limitations in obtaining an accurate description of the environment from what they perceive because they may not have all the information about the environment. The present research is focused on reinforcement learning algorithms that represent a defined category in the field of machine learning because of their unique approach based on a trial-error basis. Reinforcement learning is used to solve control problems based on received rewards. The core of its learning task is defined by a reward function where an unsuitable choice of action results in more negative rewards. The reinforcement learning framework comprises of the notion of cumulative rewards over time, to enable an agent to select actions that promote long-term results. Q-learning and SARSA are two popular methods along this approach. These two methods are similar except that Q-learning follows an off-policy strategy while SARSA is an on-policy algorithm. In this thesis, we present the comparison of Q-learning and SARSA algorithms for the global path planning of an agent in a grid-world game environment in order to verify the efficiency in different scenarios. In this thesis, simulation was performed in the grid-world environment comprising of static obstacles with a density of 30%. The results demonstrate that both approaches reach the optimal policy with a complete success rate of the learning episodes in the test cases. The comparison shows that the Q-learning algorithm outperforms the SARSA algorithm by 34% in terms of the computation time as both approaches tend closer to negative rewards while arriving at the optimal path. However, with 12% higher convergence ratio, the SARSA approach better avoids large penalties from exploratory moves which in turn proposes a safer ro

关键词： Computer science   Artificial intelligence  

摘要： Real-world problems such as human activity recognition and gesture recognition involve time series data. However, in the case of human activity recognition, for example, a machine learning model trained on one person's data may not perform well on another person's data due to different activity patterns, sensor positions, or sampling rates. This poses a problem, requiring data annotations to be collected for each person individually. Yet collecting such labels for time series may be expensive or infeasible due to the difficulty of interpreting raw time series sensor data. Instead, labels are often collected while performing each activity, but this adds to the burden of data collection. Multi-source unsupervised domain adaptation provides a remedy, by improving the performance of a machine learning model for an unlabeled (target) domain by leveraging ground truth labels in related (source) domains, i.e., leveraging data from additional people. If extra information is available about the target domain such as estimated label proportions, e.g., self-reported estimates of how many hours are spent performing various activities, then weak supervision can further improve performance through regularization. While domain adaptation has been previously studied in other contexts, prior work leaves open a number of avenues that may lead to drastically improved time series adaptation performance. We hypothesize that we can improve time series adaptation performance by (1) improving the neural network architecture used for adaptation, (2) learning a domain-invariant representation across multiple source domains and the target domain, (3) incorporating a weak supervision regularizer if target-domain label proportions are available, and (4) utilizing cross-source label information via contrastive learning. We validate each of these hypotheses through developing new adaptation models and methods that we evaluate on a variety of time series sensor datasets including human activity re

关键词： Neurosciences   Computer engineering   Computer science   Artificial intelligence   Electrical engineering  

摘要： Modeling action intentionality plays a major role in our society, with fundamental implications in the law, personal accountability and guilt. But what are the neural mechanisms that allow people to determine whether an observed behavior is intentional? Is it possible to design a machine intelligence that does the same? This dissertation presents a collection of studies to answer these questions. From human brain to machine intelligence, the studies explore possible computational mechanisms and algorithms for visual recognition and generation of intentional/non-intentional actions, as well as the regions in human brains hosting such computation. In the first study, by using machine learning algorithms on brain imaging data, we show that the brain areas associated with the visual analysis of biological motion, theory of mind, and the perception of bodies are involved in this visual recognition of intent. We demonstrate this to be the case even for minimalist, abstract scenes and inanimate agents behaving in a seemingly intentional or non-intentional manner. Furthermore, while behaviors that are easily and consistently interpreted by humans are robustly decoded in these brain regions, ambiguous ones are not, supporting a direct link between the perception of intent and these neural mechanisms. In the second and third studies, we focus on creating machine counterpart on humans' mentalizing capability, i.e., run simulations in our heads to study hypothetical actions others may perform, which, in turn, facilitates recognition of observed intentional and non-intentional actions. These two studies derives, to our knowledge, the first set of algorithms able to perform this high-level human tasks on minimalistic abstract agents (second study) and realistic human agents (third study). We then show how we can use this artificial mentalizing ability to recognize action intentionality and perform counterfactual reasoning from visual input. In the fourth study, we derive an algor

关键词： Civil engineering   Computer science   Transportation   Hurricanes   Perceptions   Floods   Households   Behavioral sciences  

摘要： The large uncertainty in population behavior with regards to how many people leave, when, from where, and to where, is a key challenge in planning hurricane evacuations. Several empirical and theoretical studies have sought to understand evacuation behavior but few have focused on developing models for predicting population evacuation behavior for future applications. In this study, we improve hurricane evacuation behavior prediction through advances in modeling and data, and offer a comprehensive evaluation of predictive power that other researchers might adopt in evaluating their models. Specifically, we modify and apply a recently introduced dynamic discrete choice framework on survey data collected in a consistent format across four hurricanes. We also take advantage of the dynamic nature of the model and include more hurricane and forecast attributes. The final set of explanatory variables can be obtained at the regional scale; hence our model can be applied in the future for prediction. Through cross validation, out-of-sample predictive power of the model is evaluated across multiple metrics, including prediction of aggregate evacuation rates, individual behavior, and evacuation timing. Cross validation results are also compared with existing statistical models from literature. At the aggregate level across the four hurricanes, the total number of evacuations was predicted with only 1% error. Individual household level results suggest 70% and 72% accurate prediction of evacuees and non-evacuees, respectively. The mean absolute departure time error is approximately only 2 hours across the population. Results across all evaluation metrics imply our model performs better when compared to existing models in the literature.

关键词： Computer science   Artificial intelligence  

关键词： Computer science   Computer engineering   Medical imaging  

摘要： Augmented Reality (AR) and Virtual Reality (VR) provide powerful, natural, and robust ways to interact with digital content, across a number of different domains. AR and VR, collectively known as Extended Reality (XR), can facilitate the execution of surgical procedures, aid in maintenance and repair of mechanical equipment, provide novel visualization paradigms for data analysis, and even empower new ways to experience video games. These experiences are built on rich, complex real-time interactive systems (RISs) that require the integration of numerous components supporting everything from rendering of virtual content to tracking of objects and people in the real world. There are decades of research on the development of robust RISs, utilizing different software engineering modalities, which facilitate the creation of these systems. While in the past, developers would frequently write all of the components and the “logical glue” themselves (often built with graphics suites such as OpenGL and DirectX), with the the rise of popular 3D game creation engines, such as Unity and Unreal, new development modalities have begun to emerge. While the underlying game engines provide a significantly easier pipeline to integrate different subsystems of AR/VR applications, there are a number of development questions that arise when considering how interaction, visualization, rendering, and application logic should interact, as developers are often left to create the “logical glue” on their own, leading to software components with low reusability. As the needs of users of these systems increase and become more complex, and as the software and hardware technology improves and becomes more sophisticated, the underlying subsystems must also evolve to help meet these needs. In this work, I present a new software design pattern, the Relay & Responder (R&R) pattern, that attempts to address the concerns found with many traditional object-oriented approaches in XR systems. The R&R pattern