教学课程资源库 更多>>

关键词： Argument mining   Computational argumentation   Computational persuasion   Computer science   Natural language processing   Persuasion  

摘要： Opinion formation and persuasion in argumentation have been shown to be affected by three major factors: the argument itself, the source of the argument, and the properties of the audience. Understanding the role of each and the interplay between them is crucial for obtaining insights regarding argument interpretation and argument generation. This is particularly important for building effective argument generation systems that can take both the discourse and the audience characteristics into account. Having such personalized argument generation systems would potentially be helpful to show individuals different viewpoints and help them make a more fair and informed decision on an issue. Even though studies in Social Sciences and Psychology have shown that source and audience effects are essential components of the persuasion process, most research in computational persuasion has focused solely on understanding the characteristics of persuasive language. In this thesis, we make several contributions to understand the relative effect of the source, audience, and language in computational persuasion. We first introduce a large-scale dataset with extensive user information to study these factors' effects simultaneously. Then, we propose models to understand the role of prior beliefs of the audience on their perception of persuasive arguments. We also investigate the role of social interactions and engagement in understanding users' success in online debating over time. We find that the users' prior beliefs and social interactions play an essential role in predicting their success in persuasion. Finally, we explore the importance of incorporating contextual information to predict argument impact and show improvements compared to encoding only the text of the arguments.

关键词： System recovery   Resource management   Petri nets   Process control   Manufacturing systems   Computer science   Tools   Multi-unit resource system (MRS)   Petri net   reconfiguration   robust deadlock control   max-controllability of siphon  

摘要： A multi-unit resource allocation system usually contains several processes and a number of resources with multiple units. Due to the competition for shared resources in these systems, deadlocks may occur. Recently, researchers have shown an increased awareness in deadlock control strategies for such a kind of systems without considering the dynamic changes such as processing failures and rework by using the Petri net paradigm. This article reports a new strategy for deadlock analysis and control in reconfigurable multi-unit resource systems (MRSs). We discuss a generalized class of Petri nets in which each stage of a process may require a number of units of different types of resources to perform a task. In this way, we can model more complex real systems. Thanks to a generalized class of Petri nets, i.e., the system of sequential systems with shared resources ((SR)-R-4), this article proposes an effective integrated strategy for designing robust supervisors for reconfigurable MRSs, and improves an (SR)-R-4 model to achieve a new model, namely a system of sequential systems with shared resources and part-re-entry ((SRP)-R-4), which represents the procedure that a flawed product re-enters a system and is re-processed. We use a siphon-based max-controllability deadlock prevention policy (DPP) to supervise the evolution of the (SRP)-R-4, and present a comprehensive analysis to demonstrate that the controlled (SRP)-R-4 is free of deadlocks. A net analysis tool (INA) is used to test and validate the resulting (SRP)-R-4.

关键词： Computer science   Computer engineering   Artificial intelligence   Information technology  

摘要： Cloud providers need to optimize the container deployments to efficiently utilize their network, compute and storage resources. In addition, they require an attractive pricing strategy for the compute services like containers, virtual machines, and serverless computing in order to attract users, maximize their profits and achieve a desired utilization of their resources. This thesis aims to tackle the twofold challenge of achieving high performance in container deployments and identifying the pricing for compute services. For performance, the thesis presents a transport-adaptive network architecture (D-TAIL) improving tail latencies. Existing transport protocols such as Homa, pFabric [1, 2] utilize Shortest Remaining Processing Time (SRPT) scheduling policy which is known to have starvation issues for long flows as SRPT prioritizes short flows. D-TAIL addresses this limitation by taking age of the flow in consideration while deciding the priority. D-TAIL shows a maximum reduction of 72%, 29.66% and 28.39% in 99th-percentile FCT for transport protocols like DCTCP, pFabric and Homa respectively. In addition, the thesis also presents a container deployment design utilizing peer-to-peer network and virtual file system with content-addressable storage to address the problem of cold starts in existing container deployment systems. The proposed deployment design increases compute availability, reduces storage requirement and prevents network bottlenecks. For pricing, the thesis studies the tradeoffs between serverless computing (SC) and traditional cloud computing (virtual machine, VM) using realistic cost models, queueing theoretic performance models, and a game theoretic formulation. For customers, we identify their workload distribution between SC and VM to minimize their cost while maintaining a particular performance constraint. For cloud provider, we identify the SC and VM prices to maximize its profit. The main result is the identification and characterization of th

关键词： Decomposition   Principal components analysis   Algorithms   Colorectal cancer   Genotype & phenotype   Neighborhoods   Neural networks   Artificial intelligence   Computer science  

摘要： Viewed at a high spatial and molecular resolution, every individual instance of a biological tissue is unique. However, it is a tenet of evolution that prior evolutionary successes are adapted and repurposed for variations on function and for the creation of new functions. Thus, it would be expected that the organization characteristic of a tissue’s type can be expressed in terms of a collection of repeated biological units that respect some rules governing their assembly and collective functionality. Advances in high-parameter imaging technologies present the opportunity to observe tissues at high spatial and molecular resolution. What, then, are the repeated biological units and the rules (governing the units’ assembly and collective functionality), characterizing the organization of tissue types viewed through the lens of such technologies? We provide conceptual, mathematical and algorithmic tools towards possible answers of this question, and apply them to high-parameter tissue imaging data of lymphoid tissues as well as the immune-tumor microenvironment. The results yield clinically relevant insights into the specific biology of these tissues, as well as hint at general principles of tissue organization that become apparent only when tissues are assayed in such detail. We start considering cell types as the repeated units, describing those in lymphoid tissues with a disentangled representation learning strategy as well as in terms of cellular contacts. However, interpreting the outputs of a weakly-supervised neural network trained on such data suggested that specific regions of the spleen could be driving pathology in a murine model of autoimmunity. Urban neighborhoods (like the business district or residential area), are not merely regions, but instead have collective behaviors characteristic of cities. We therefore define cellular neighborhoods (CNs) and their collective behaviors (modular organization, functional states and communication) to describe tissues

关键词： Computer science   Artificial intelligence  

摘要： Modern unmanned aerial vehicles have become more sophisticated from their earlier 20th century predecessors. Using unmanned aerial vehicles (UAV), or drones, in tasks involving navigating through unknown surroundings, such as wildfire monitoring, target tracking, or search and rescue, is becoming more widespread. They can host a wide range of sensors to measure in an environment with relatively low operating costs and high flexibility. The problem is that most current research tries to rely on the accuracy of the drone model to describe the target or prior knowledge of the environment. The implementation of a UAV will offer a way to accomplish tasks in unknown surroundings. Using a reinforcement learning (RL) algorithm, the drone can navigate efficiently from an arbitrary starting position to a goal position in the shortest possible way. The main contribution is to provide a framework for applying an RL algorithm to enable UAV to operate in such an environment. To carry out the given task, the UAV must have a component to make it learn to allow it to find the way to the goal in an optimal fashion. The simulation and real implementation to indicate how the UAVs can successfully learn to navigate through an unknown environment will be conducted.

关键词： Computer Engineering   Computer science   Electrical engineering   Artificial intelligence  

摘要： Internet of drones (IoD), which utilize drones as Internet of Things (IoT) devices, deploys several drones in the air to collect ground information and send them to the IoD gateway for further processing. Computing tasks are usually offloaded to the cloud data center for intensive processing. However, many IoD applications require real-time processing and event response (e.g., disaster response and virtual reality applications). Hence, data processing by the remote cloud may not satisfy the strict latency requirement. Fog computing attaches fog nodes, which are equipped with computing, storage and networking resources, to IoD gateways to assume a substantial amount of computing tasks instead of performing all tasks in the remote cloud, thus enabling immediate service response. Fog-aided IoD provisions future events prediction and image classification by machine learning technologies, where massive training data are collected by drones and analyzed in the fog node. However, the performance of IoD is greatly affected by drones' battery capacities. Also, aggregating all data in the fog node may incur huge network traffic and drone data privacy leakage. To address the challenge of limited drone battery, the power control problem is first investigated in IoD for the data collection service to minimize the energy consumption of a drone while meeting the quality of service (QoS) requirements. A PowEr conTROL (PETROL) algorithm is then proposed to solve this problem and its convergence rate is derived. The task allocation (which distributes tasks to different fog nodes) and the flying control (which adjusts the drone's flying speed) are then jointly optimized to minimize the drone's journey completion time constrained by the drone's battery capacity and task completion deadlines. In consideration of the practical scenario that the future task information is difficult to obtain, an online algorithm is designed to provide strategies for task allocation and flying control when

关键词： Internet of Things   Malware   Neural networks   Support vector machines   Artificial intelligence   Computer science  

摘要： Although desktops and laptops have historically composed the bulk of botnet nodes, Internet of Things (IoT) devices have become more recent targets. Lightbulbs, outdoor cameras, watches, and many other small items are connected to WiFi and each other; and few have well-developed security or hardening. Research on botnets typically leverages honeypots, PCAPs, and network traffic analysis tools to develop detection models. The research questions addressed in this Problem Report are: (1) What machine learning algorithm performs the best in a binary classification task for a representative dataset of malicious and benign IoT traffic; and (2) What features have the most predictive power? This research showed that the best performing algorithms were Random Forest with accuracy of 97.45% and F1 score of 97.39%; and the Linear SVM with a recall score of 99.90%. The most important features for the classification were: time of day, history, protocol, and count of origin bytes sent. Of these, time of day and volume of traffic coming from the same IP addresses are consistent for port scanning, infection, and distributed denial of service attacks.

关键词： Clothing   Software   Armed forces   Computer science   Textile research  

摘要： Moving while wearing garments, especially protective garments, increase their wearer’s energy expenditure thereby lowering performance and endurance. To lower the burden of garments, design and fit elements are employed to optimize ease of motion, but precisely measuring the effects of these design features remains difficult. Currently employed measurement methods often fail to distinguish between garments, which do not greatly differ in material or end use. More precise measurement techniques require equipment and resources that make them difficult to incorporate into garment development timelines. 3D garment simulation has been increasingly incorporated into the garment development processes to increase speed and may also be able to address ease of motion measurements through the quantitative fit analysis tools enabled by the technology. This project explores the ability of 3D garment simulation to measure ease of motion in garments which share materials but differ in design and fit. To accomplish this, two garment designs were developed, one matching the current design of the army combat uniform (ACU) jacket, and a jacket developed using a novel pattern drafting technique where body measurements are taken from multiple poses and composited into a single pattern. Instead of relying on conventional ease recommendations, the novel technique defined garment ease based on the maximum body dimensions achieved during motion and created a pattern with less circumferential ease, but greater vertical ease in the shoulder area. Both designs were evaluated using a traditional wear trial involving live subjects who performed calisthenic exercises and subjectively evaluated comfort and ease of motion. These results were compared with a simulated wear trial using stretch mapping in the Optitex program. In the physical wear trial, the experimental garment was rated as less comfortable and less preferred with the shoulders and chest area being the most problematic. The simulated

关键词： Algorithms   Neural networks   Artificial intelligence   Computer science  

摘要： Reinforcement learning has shown tremendous success over the past few years. Much of this recent success can be attributed to agents learning from an inordinate amount of data in simulated environments. In order to achieve similar success in real environments, it is crucial to address data efficiency. Uncertainty quantification plays a prominent role in designing an intelligent agent which exhibits data efficiency. An agent which has a notion of uncertainty can trade-off between exploration and exploitation and explore in an intelligent manner. Such an agent should not only consider immediate information gain from an action but also its consequences on future learning prospects. An agent which has this capability is said to exhibit deep exploration. Algorithms that tackle deep exploration, so far, have relied on epistemic uncertainty representation through ensembles or other hypermodels, exploration bonuses, or visitation count distributions. An open question is whether deep exploration can be achieved by an incremental reinforcement learning algorithm that tracks a single point estimate, without additional complexity required to account for epistemic uncertainty. We answer this question in the affirmative. In this dissertation, we develop Langevin DQN, a variation of DQN that differs only in perturbing parameter updates with Gaussian noise, and demonstrate through a computational study that Langevin DQN achieves deep exploration. This is the first algorithm that demonstratively achieves deep exploration using a single-point estimate. We also present index sampling, a novel method for efficiently generating approximate samples from a posterior over complex models such as neural networks, induced by a prior distribution over the model family and a set of input-output data pairs. In addition, we develop posterior sampling networks, a new approach to model this distribution over models. We are particularly motivated by the application of our method to tackle reinforcem

关键词： Computer science   Artificial intelligence   Electrical engineering  

摘要： Efficient hardware, increased computational power, and smart sensors, are powering deep learning, and moving intelligence from the cloud to edge devices (e.g., smartphones, smart cameras, and wearables). However, deep neural networks are computationally expensive and are difficult to deploy on edge devices because of limited computational capabilities, including limited energy overhead. To enable on-device AI, we focus on developing efficient neural architectures for edge devices. We optimize the basic building blocks of deep neural networks (e.g., convolutional layers and linear transformation functions) and build light-weight, fast, and memory-efficient deep neural architectures. Besides efficiency, we also study the generalization capabilities of our architectures on different datasets and tasks. We start by designing efficient architectures for computer vision tasks. In the first part of the dissertation, we introduce the Efficient Spatial Pyramid (ESP) unit, an efficient alternative to standard convolution layers in convolutional neural networks (CNNs). Compared to standard convolutions, the ESP unit allows networks to learn representations from a large receptive field with fewer parameters and operations. Our efficient architecture, ESPNet, that is built using the ESP module is able to deliver similar or better performance than state-of-the-art efficient neural architectures, such as MobileNets and ShuffleNets, across different tasks (e.g., image classification and object detection) while being 2 times more power efficient. The second part is geared towards improving the performance of efficient CNNs. We introduce a novel and generic convolutional unit, the DiCE unit, that is built using dimension-wise convolutions and dimension-wise fusion. The dimension-wise convolutions apply light-weight convolutional filtering across each dimension of the input tensor, while dimension-wise fusion efficiently combines these dimension-wise representations; allowing the DiCE