教学课程资源库 更多>>

关键词： Computer science  

摘要： Deep convolutional neural networks (CNNs) trained on visual objects have shown intriguing ability to predict some response properties of visual cortical neurons. However, the factors (e.g., if the model is trained or not, receptive field size etc.) and computations (e.g., convolution, rectification, pooling, normalization etc.) that give rise to such ability, at what level, and the role of intermediate processing stages in explaining changes that develop across areas of the cortical hierarchy, are poorly understood. We focused on the sensitivity to textures as a paradigmatic example, since recent neurophysiology experiments provide rich data pointing to texture sensitivity in secondary (but not primary) visual cortex (V2). We initially explored the CNN without any fitting to the neural data, and found that the first two layers of the CNN showed qualitative correspondence to the first two cortical areas in terms of texture sensitivity. After we have shown a qualitative similarity between the CNN models and the brain V2 data, we sought to develop a approach quantify this correspondence. We developed methods to select same number of CNN model neurons that best fits the brain neural recordings. We found that the CNN could develop compatibility to secondary cortex in the second layer following rectification, and that this was improved following pooling, but only mildly influenced by the local normalization operation. Higher layers of the CNN could further, though modestly, improve the compatibility with the V2 data. The compatibility was reduced when incorporating random rather than learned weights. We also considered several other popular models in the literature such as Hmax and ScatNet. Hmax is a model of visual cortex that builds an increasingly complex and invariant feature representation in the higher layers. ScatNet is a specialized model for texture discrimination by incorporating higher order moments. We found that only the second layer in both the models has so

关键词： Communication   Computer engineering   Computer science  

摘要： In this thesis, several Collaborative Filtering (CF) approaches with latent variable methods were studied using user-item interactions to capture important hidden varia- tions of the sparse customer purchasing behaviours. The hidden latent factors are used to generalize the purchasing pattern of the customers and provide product recommenda- tions. CF with Neural Collaborative Filtering (NCF) was shown to produce the highest NDCG performance on the proprietary dataset provided by the Company. Different hy- perparameters were tested for applicability in the CF framework. External data sources like click-data and metrics like Clickthrough Rate (CTR) were reviewed for potential extensions to the work presented. The work presented in this thesis provides techniques, the Company can use to provide product recommendations to enhance revenues, attract new customers, and gain advantages over competitors.

关键词： Computer science  

摘要： Data Communication has two fundamental aspects. First is moving data from a locality of a logical endpoint to a desired destination. Secondly are the enabling methods of notification or receipt of data at respective logical endpoints. While optimizing communication can be trivially defined as both the transferring and notification of data by the shortest path possible, it is, practically speaking, a difficult problem because of the existence of many performance inhibitors. This thesis will show that communications performance is deeply affected by the behaviour of the underlying execution environment. High performance runtime execution environments broadly expose two types of communication, namely, two-sided and one-sided. Two-sided communication actively involves both source and target processors, but may incur additional overheads. One-sided communication defines messaging patterns where data transfer does not involve any of the processing components of the destination. Cutting edge networks interconnect technologies offer hardware assisted one-sided communication primitives such as Remote Direct Memory Access (RDMA) - A capability that minimizes remote memory access latency by bypassing the remote host processor and avoiding operating system interaction. While RDMA is extremely useful, exposing proper abstraction and application are of major concern because transfer and notification paths need to be handled separately, but cautiously. This thesis will show that communication performance can be improved by building lightweight network software architectures that can be seamlessly integrated by both the hardware and software interface. More specifically for optimizing communication between nodes (i.e. inter-node), the thesis introduces a novel distributed dynamic remote memory allocator for RDMA called “rmalloc”, and a high performance communication channel for RDMA, called “rpipe”. The “rmalloc” and “rpipe” introduce a light weight remote memory provisioning capab

关键词： Computer science  

摘要： We investigate two topics from computational differential geometry: quadratic differentials and spherical optimal transportation. In the first part, we investigate the surface foliation and quadratic differentials. A foliation decomposes the surface into a family of closed loops, such that the decomposition has local tensor product structure. By projecting each loop to a point, the surface is collapsed into a graph. We compute the harmonic map from surface to the graph. The Hopf differential of this harmonic map is a quadratic differential, which induces a surface foliation. The foliation structure and quadratic differential provides plentiful information for potential applications. The quadratic differentials on surfaces form a linear space, which plays a fundamental role in Teichmüller theory. We propose a novel model of the space of landmark constrained diffeomorphisms. Based on Teichmüller theory, this mapping space is generated by the Beltrami coefficients that are infinitesimally Teichmüller equivalent to zero. By using this theoretical model, optimal registrations can be achieved by iterative optimization with linear constraints in the diffeomorphism space, such as harmonic maps and Teichmüller maps, which minimize different types of distortions. The theoretical model is rigorous and has practical value. Our experiment results demonstrate the efficiency and efficacy of the proposed method. In the second part, we extend a planar discrete optimal transportation theory and algorithms to spherical case. Optimal transportation (OT) problem aims at finding the most economic way to transform one probability measure to the other, which plays a fundamental role in many fields, such as computer graphics, computer vision, machine learning, geometry processing and medical imaging. Most existing algorithms focus on searching the optimal transportation map in Euclidean space, based on Kantorovich theory or Brenier theory. This work introduces a novel theoretical framework

ISBN: (纸本)9787115523648

关键词： Clustering   Generative Adversarial Network   Latent factor models   Matrix completion   Mutual information   Unsupervised learning   Computer science   Artificial intelligence   To Be Assigned  

摘要： Unsupervised learning is the branch of machine learning that is aimed at learning patterns from data without labels. Supervised learning with millions of labels for image classification had driven the modern deep learning revolution in the past few years. Deep neural networks have exceeded human performance at this specific task. But the requirement of such large amounts of labeled data for these models makes one skeptical about the generalization of such intelligence to myriad tasks. While training a neural network to classify images of a cat, one might wonder : Do humans really need hundreds of images to differentiate a cat from an elephant ? Or is there some underlying principle that can be rendered useful by a machine in its race to match human intelligence ? Unsupervised learning unveils the potential of machine learning algorithms beyond empirical risk minimization and extend them to learning non-trivial representations of the data. At the core of such learning are two distinct principles - model-agnostic representation learning and model-guided inference. The goal of this thesis is to extend the present literature on unsupervised learning through design of novel unsupervised algorithms for clustering, information estimation and model-guided inference. Our journey starts with one of the simplest, yet most fundamental unsupervised learning problem, namely clustering. We explore how modern generative principles such as Generative Adversarial Learning (GAN) can be used to cluster diverse types of data. Even though auto-encoders had been used for clustering in the past, clustering using GANs was unexplored prior to this work. ClusterGAN modifies the vanilla GAN architecture to enable embedding of data in the latent space where cluster structure is revealed. It also improves the generation ability of vanilla GANs by segregating a complex multi-modal distribution into simpler components. Recently, information-theoretic quantities such as mutual information and cross

关键词： Educational technology   Computer science  

摘要： In the last decade, the immense growth of computational power, enhanced data storage capabilities, and the increasing popularity of online learning systems has led to adaptive learning systems becoming more widely available. Parallel to infrastructure enhancements, more researchers have started to study the adaptive task selection systems, concluding that suggesting tasks appropriate to students' needs may increase students' learning gains. This work built an adaptive task selection system for undergraduate organic chemistry students using a deep learning algorithm. The proposed model is based on a recursive neural network (RNN) architecture built with Long-Short Term Memory (LSTM) cells that recommends organic chemistry practice questions to students depending on their previous question selections. For this study, educational data were collected from the Organic Chemistry Practice Environment (OPE) that is used in the Organic Chemistry course at Arizona State University. The OPE has more than three thousand questions. Each question is linked to one or more knowledge components (KCs) to enable recommendations that precisely address the knowledge that students need. Subject matter experts made the connection between questions and related KCs. A linear model derived from students' exam results was used to identify skilled students. The neural network based recommendation system was trained using those skilled students' problem solving attempt sequences so that the trained system recommends questions that will likely improve learning gains the most. The model was evaluated by measuring the predicted questions' accuracy against learners' actual task selections. The proposed model not only accurately predicted the learners' actual task selection but also the correctness of their answers.

关键词： Educational leadership   Women's studies   Management  

摘要： The purpose of this case study was to understand the experiences of women in any leadership role with Computer Science or Engineering degrees that have helped them persist and advance in their careers in the context of Silicon Valley. The “leaky pipeline” phenomenon has been described as women leaving the Science, Technology, Engineering, and Mathematics (STEM) track at many stages in the pipeline (K-12, secondary, career). It has resulted in underrepresentation of women at all levels of the career ladder, specifically, for women with Computer Science and Engineering degrees in leadership positions. Using a case study research methodology, this study sought to answer this question: Why do women with Computer Science and Engineering degrees leadership positions in Silicon Valley stay in the field? Ten women in senior leadership roles in high-tech companies in Silicon Valley with at least a Bachelor's in Science degree in Computer Science or Engineering participated in the study. Women in this case study had a job title of Director or above. One-on-one interviews, a review of artifacts, and the researcher's observations of the participants during their participation in the study were analyzed. Four themes emerged from their interviews and artifact review: (a) STEM foundation, (b) grit, (c) Silicon Valley barriers, and (d) career strategies.

关键词： Computer science  

摘要： With the rise in software ecosystem initiatives, developing Application Programming Interfaces (APIs) has become an increasingly common practice. One main concern in developing APIs is that they expose back-end systems and data towards clients. This exposure threatens critical non-functional requirements, such as security of the back-end systems, performance of the provided services, and privacy of the communications with clients. Although dealing with non-functional requirements during software design has been long studied, there is still no framework to specifically assist software developers with addressing these requirements in APIs. In this thesis, we introduce Rational API Designer (RAPID), an assistant that provides consultation about designing non-functional requirements in the architecture of APIs. We have equipped RAPID with a broad range of expert knowledge about API design, systematically collected and extracted from the literature. The API design knowledge has been encoded as a set of 156 catalogues using the Non-Functional Requirements (NFR) language, a multi-valued logic commonly used for describing non-functional and functional requirements of software systems. RAPID uses the API design catalogues in a step-wise refinement and selection procedure to arrive from a given requirement, to a set of design alternatives, to a final suggestion for a given API design specification. Seven well-experienced developers have blindly evaluated accuracy of RAPID’s consultations over seven different cases of API design and on providing design guidelines for thirty design questions. The results of evaluation show that RAPID’s recommendations meet acceptable standards of the majority of the evaluators 73.3% of the time. Moreover, analysis of evaluators’ comments suggests that more than one-third of the unacceptable ratings (34%) given to RAPID’s answers are due to valid but incomplete design guidelines. We thus expect that the accuracy of the consultations will increas

关键词： Information science   Computer science  

摘要： The aim of a question-answering (QA) system is to provide answers to users’ questions automatically. Depending on the scope of the topics of the input questions, the general QA is categorized into open-domain QA and closed-domain QA. Open-domain QA is a system that takes as input a question in natural language and an external information source, e.g., text from an encyclopedia or a knowledge base, and returns the answer if it is expressed in that source. Finding the answer to an arbitrary question on any topic is a complex and challenging task. The challenging nature of open-domain QA motivates work on two feasible versions of the general QA problem: community Question Answering (cQA) for open-domain QA and a new QA task on medical data for closed-domain QA. cQA websites enable users to ask questions and contribute answers. One useful strategy for minimizing the response latency between the submission of questions by users and the subsequent posting of answers is to search the QA repository for similar questions that have already been answered. For open-domain QA, I addressed the task of question ranking in cQA: ranking previously answered questions in a question repository with respect to their relevance to a new, unanswered reference question. Neural Sequence Learning with Attention (SLA) is proposed to perform similarity-based ranking of questions that have already been answered. For closed-domain QA, a new question-answering paradigm over medical data is proposed in which users query the system using both natural language and direct interactions (mouse clicks) within a Graphical User Interface called PhysioGraph. Doctors access data acquired from patients with type I diabetes using PhysioGraph, which displays sensor data as time series one day at a time. To interact with PhysioGraph, doctors rely on traditional modes of interaction, such as the mouse or the keyboard, which are slow and inefficient. Enabling doctors to interact with PhysioGraph in natural languag