教学课程资源库 更多>>

关键词： Computer engineering   Electrical engineering   Computer science  

摘要： The ankle joint complex is a common source of injury for various demographics and is often observed during gait analysis. I investigate using soft robotic sensors as a means for collecting kinematic data at the ankle joint complex. I validate the linearity of these sensors by measuring stretch against extension and against stretch from frontal and sagittal planar foot movements using a wooden ankle mockup. I then conduct a study involving ten participants who perform repetitive trials of four foot movements (plantarflexion, dorsiflexion, inversion and eversion) using ten different locations. Four optimal locations were identified for these movements based on linearity, accuracy, robustness, and consistency. Lastly, I validated soft robotic sensors against the human gait cycle. Twenty participants were recruited and performed twelve trials, walking across a flat surface and a cross-sloped surface while motion capture data and soft robotic sensor data was collected.

关键词： Robotics   Computer science  

摘要： Earlier work in the field of intent recognition used laser and cameras to track people and extract physical information to train and test models of intention. With the progression of computation abilities, Neural Networks have made it possible to extract additional information that earlier work was not able to take advantage of, primarily person pose information. With this new information, intent recognition systems should be able to differentiate in finer detail between neighboring intents. We combine all the available information for both pose and movement descriptors to generate profiles about each person in the scene. This new pose information also makes it possible to track people in laser data more reliably. To do this tracking well, calibration between the sensors in critical, so we propose an iterative method for calibration that can produce a result not just for robots with sensors aligned toward the face of the target but also for sensor arrays that may observe the target from disparate directions. The calibration method we developed is only one of just a few current methods that can solve the calibration of a laser range finder and an RGB camera with only a single sample and unknown target dimensions.

关键词： Information technology   Computer science  

摘要： Since the dawn of civilization, we have striven to invent tools to support our productivity and creativity. A key role of these tools is to maintain, display, and provide means to operate on certain representations of information. The ideal representations visualize information in forms that can be perceived effortlessly and afford interactions which enable users to reason, analyze, and make sense of the information. Perhaps the most complex tool invented so far is digital computer. Since its inception, research has been striving to leverage computation to augment our productivity and creativity. The increasingly fast computation has shifted the bottleneck of employing computation from the hardware specifications to the representation of and interaction with digital information. The Window, Icons, Menus, Pointer (WIMP) user interface has been the dominant Graphical User Interface (GUI) framework of digital computers for 50 years. The many problems of the WIMP have long been recognized and yet unresolved: it relies on tedious manipulation of complex interface elements to achieve simple tasks; with mouse and keyboard as primary input, it does not effectively leverage the rich interaction capabilities we possess. We contribute a novel representation, the Object-Oriented Representation. By representing abstract elements in users’ workflow as concrete objects, the Object-Oriented Representation offloads abstract content and structure from users’ minds to external interfaces, which reduces cognitive load. By directly matching users’ internal representations of the information, it enables users to directly articulate their intentions, rather than translating them into tedious actions constrained by inappropriate representations. The smallest unit of GUI, attribute, is first objectified as Attribute Object. Attribute Object affords rich interactions which were tedious or impossible, and allows for the composition of higher-level elements to address other complex tasks, incl

关键词： Artificial intelligence   Computer science  

摘要： Decision-making based on probabilistic reasoning typically involves selecting a subset of several expensive observations that best predict the system state. The contribution of an observation to the overall utility of the system is referred to as its value of information (VoI). Non-myopically (non-greedily) choosing the optimal observations is wildly intractable even in simple probabilistic graphical models. This work includes four distinct contributions. First, we present a framework, based on the Markov Decision Process (MDP), for optimally choosing observations in general dynamic Bayesian networks (DBNs), based on non-myopic VoI. The framework involves a definition of utilities which incorporates a wide range of information metrics, and uses a filtering strategy of choosing observations. When selecting observations in chain graphical models, the exact algorithms resulting from the framework offer significant complexity improvements over the current state-of-the-art. Second, we use the MDP-based framework to formulate efficient approximation algorithms that offer absolute, deterministic, a posteriori approximation bounds. The algorithms come with a parameter which can be tuned to adjust the amount of computational resources used. We show that affording more computational resources lead to better results, and tighter approximation bounds. We leverage this fact to create useful anytime algorithms. Third, the preceding algorithms, as well as all other works that have used the generic notion of VoI considered here, require a graphical model with a finite number of variables. We present another MDP-based algorithm for exactly optimizing non-myopic VoI in a restricted class of graphical models of unbounded size. We also discuss how the related concepts can be integrated into the preceding approximation algorithms in order to make them compatible with general graphical models of unbounded size. And last, we present a setting for optimally choosing observations, based on

关键词： Artificial intelligence   Computer science  

摘要： Neural Processes (NPs) are a class of regression models that learn a map from a set of input-output pairs to a distribution over functions. NPs are computationally tractable and provide a number of benefits over traditional nonlinear regression models. Despite these benefits, there are two main domains where NPs fail. This thesis is focused on presenting extensions of the Neural Process to these two areas. The first of these is the extension of Neural Processes graph and network data which we call Graph Neural Processes (GNP). A Graph Neural Process is defined as a Neural Process that operates on graph data. It takes spectral information from the graph Laplacian as inputs and then outputs a distribution over values. We demonstrate Graph Neural Processes in edge value imputation and discuss benefits and drawbacks of the method for other application areas. The second extension of Neural Processes comes in the fundamental training mechanism. NPs are traditionally trained using maximum likelihood, a probabilistic technique. We show that there are desirable classes of problems where NPs fail to learn. We also show that this drawback is solved by using approximations of the Wasserstein distance. We give experimental justification for our method and demonstrate its performance. These Wasserstein Neural Processes (WNPs) maintain the benefits of traditional NPs while being able to approximate new classes of function mappings.

关键词： Task analysis   Planning   Resource management   Trajectory   Sea surface   Computer science   Ant colony optimization (ACO)   autonomous robotics   autonomous underwater vehicle (AUV)   differential evolution (DE)   target traveling mission  

摘要： Multiple autonomous underwater vehicles (AUVs) are popular for challenging submarine missions. In this article, we focus on the multi-AUV path planning for a common class of missions that need to traverse lots of mission targets in large and complex environments. Given that AUVs are often launched from a movable surface vehicle, e.g., a ship, a multi-AUV target traveling problem is formulated with a requirement of surface point location. Thereafter, a hybrid metaheuristic approach is developed by sequentially performing a cube-based environment modeling, cost map building, voyage planning, and detailed trajectory planning. Specifically, the shortest path faster algorithm (SPFA) is adopted to build a cost map among targets and candidate surface points, and the A* search is utilized for trajectory planning. The optimality of both SPFA and A* can indeed be guaranteed. Thus, voyage planning becomes critical and an algorithm namely DE-C-ACO is proposed by combining the ant colony optimization (ACO) and the differential evolution (DE) with a cluster-based adjustment strategy, i.e., DE-C. DE-C and ACO evolve in parallel for surface point location and voyage generation, respectively. Experiments based on realistic bathymetries are conducted and the results validate the effectiveness and efficiency of the proposed DE-C-ACO.

关键词： Computer science  

摘要： Video Understanding is a fundamental task for computer vision and artificial intelligence (AI). Human’s visual system can rapidly recognize different objects and their status while in a dynamic process and support us to interact with the environment in real-time. The representative applications are autopilot, sports analysis, and human-robot interaction. Convolutional Neural Networks (CNNs) have gained great success in tasks such as image classification and object detection. Most of the previous work has focused on still images, but in the real world, changes never stop, it is necessary to extend the detection tasks from static images to videos. Different from still images, to understand the content of a video, a computer needs to understand the changes of all the objects this video contains along the temporal dimension, make clear of their relationship, and then infer what happens in this video. For instance, with a still image, if there is a person standing in front of a chair, he might just stand up, or is going to sit on it. It is hard to guess what the real action is. However, with the help of temporal information, if this person is sitting on the chair in the following several frames, it is not difficult to infer that this is a sit-down action. Following this idea, in this dissertation, we explore video understanding with two detection tasks: 1) video object detection, which focuses on localizing and categorizing objects in videos while their appearances and locations may change in continuous frames and 2) action detection, which aims to localize where the actions happen, and categorize which actions they are. It needs to consider not only the relationship of each object contains in the video, such as people and football, but also how that relationship changes across frames. We explore video object detection with the polyp localization task in colonoscopy video. We propose a two-stream structure to detect polyps with both spatial and temporal representation and then

关键词： Educational technology   Information science   Computer science  

摘要： Games are hypothesized to be an effective alternative for training than other methods. Prior research showed that learners often find training boring, and when they took training with games, they reported higher engagement, motivation, and a positive perception of the learning experience. The hypothesis for this study was that participants who take game-based information security awareness training would perform statistically significantly better than participants who took video-based training. One hundred participants were given a pretest and posttest with half of the participants using video-based information security awareness training, and the other participants using game-based information security awareness training. Conducting data analysis using IBM SPSS Statistics 24, it was discovered that the group receiving the video-based training performed significantly better on the posttest and had a higher mean score than the video-based training group.

关键词： Electrical engineering   Computer science  

摘要： Image denoising is an important problem in image processing and computer vision. In real-world applications, denoising is often a pre-processing step (so-called low-level vision task) before image segmentation, object detection, and recognition at higher levels. Traditional image denoising algorithms often make idealistic assumptions with the noise (e.g., additive white Gaussian or Poisson). However, the noise in the real-world images such as high-ISO photos and microscopic fluorescence images are more complex. Accordingly, the performance of those traditional approaches degrades rapidly on real-world data. Such blind image denoising has remained an open problem in the literature. In this project, we report two competing approaches toward blind image denoising: supervised and unsupervised learning. We report the principles, performance, differences, merits, and technical potential of a few blind denoising algorithms. Supervised learning is a regression model like CNN with a large number of pairs of corrupted images and clean images. This feed-forward convolution neural network separates noise from the image. The reason for using CNN is its deep architecture for exploiting image characteristics, possible parallel computation with modern powerful GPU’s and advances in regularization and learning methods to train. The integration of residual learning and batch normalization is effective in speeding up the training and improving the denoising performance. Here we apply basic statistical reasoning to signaling reconstruction to map corrupted observations to clean targets Recently, few deep learning algorithms have been investigated that do not require ground truth training images. Noise2Noise is an unsupervised training method created for various applications including denoising with Gaussian, Poisson noise. In the N2N model, we observe that we can often learn to turn bad images to good images just by looking at bad images. An experimental stud

关键词： Computer science  

摘要： This paper analyzes the main approaches to automated information collection on the Internet, and provides a description of the context for collecting information. A system for automatically collecting information (parsing) on the network in the Golang programming language. © 2020, Journal of Computing and Information Technology. All rights reserved.