教学课程资源库 更多>>

关键词： Computer science  

摘要： Over the last few years, advances have been made in fields like scalable algorithms, novel learning models, and machine learning software frameworks, which resulted in further breakthroughs in computer vision, natural language processing, and recommendation systems. In entering the era of big data, large scale machine learning became increasingly important in training a big model on a big dataset. However, even with modern distributed systems, the large amount of data used today cannot be fully loaded into memory or manipulated directly. Stochastic optimization algorithms can reduce the size of model and data in learning process, therefore plays a key role in building a large scale machine learning system. Modern machine learning faces two kinds of big data: (1) the number of features is much larger than the sample size, in other words, high-dimension; (2) massive sample size. The two types of data need different treatments. For (1), the challenge lies in finding the intrinsic structure of the model used. Norm regularized estimators like Lasso is widely used in finding a low complex parameter with high statistical accuracy. More complicated structures can be constructed using simple structures like sparsity and low-rank, which are shown to be more robust than single structured models. For (2), carefully designed algorithms that make use of only a small portion of data in each operation are necessary for computation. Stochastic gradient descent (SGD) and its variants are well known for their practical effectiveness. Prior information can be incorporated in algorithms to improve their efficiency while maintaining convergence guarantee. In this thesis, we target a unified understanding of stochastic algorithms in both (1) and (2). For high-dimensional data, we consider estimation from superposition structure and locally low-rank structure. Both structures are constructed from simple structures like sparsity and low-rankness. A parameter is superposition structured if i

关键词： Computer Science   Engineering   Deep neural networks   Supervised learning   Attention   Speech enhancement   Speech denoising   Speech dereverberation   Time-frequency masking   Speech intelligibility   Speech quality   Computational auditory scene analysis  

摘要： In daily listening environments, the speech reaching our ears is commonly corrupted by both room reverberation and background noise. These distortions can be detrimental to speech intelligibility and quality, and also pose a serious problem for many speech-related applications, including automatic speech and speaker recognition. The objective of this dissertation is to enhance speech signals distorted by reverberation and noise, to benefit both human communications and human-machine interaction. Different from traditional signal processing approaches, we employ deep learning approaches to perform reverberant-noisy speech enhancement. Our study starts with speech dereverberation without background noise. Reverberation consists of sound wave reflections from various surfaces in an enclosed space. This means the reverberant signal at any time step includes the damped and delayed past signals. To explore such relationships at different time steps, we utilize a self-attention mechanism as a pre-processing module to produce dynamic representations. With these enhanced representations, we propose a temporal convolutional network (TCN) based speech dereverberation algorithm. Systematic evaluations demonstrate the effectiveness of the proposed algorithm in a wide range of reverberant conditions. Then we propose a deep learning based time-frequency (T-F) masking algorithm to address both reverberation and noise. Specifically, a deep neural network (DNN) is trained to estimate the ideal ratio mask (IRM), in which the anechoic-clean speech is considered as the desired signal. The enhanced speech is obtained by applying the estimated mask to the reverberant-noisy speech. Listening tests show that the proposed algorithm can improve speech intelligibility for hearing-impaired (HI) listeners substantially, and also benefit normal-hearing (NH) listeners. Considering the different natures of reverberation and noise, we propose to perform speech enhancement using a two-stage strategy,

关键词： Computer science  

摘要： The amount of data being produced every day is growing at a very high rate, opening the door to new knowledge while also bringing forth cyber breach opportunities for malicious users. In this thesis, the objective is to analyze public data to gain valuable insight for cybersecurity applications. Using public Twitter account data, a machine learning model is trained to identify bot accounts which helps lower the amount of fake news and malicious users. A survey of text summarization techniques to identify the best method for summarizing public data in the domain of cybersecurity is presented. A web application is also created to serve as a public tool for users to summarize input text of their choosing using a variety of algorithms. The contribution of this thesis is thus twofold: a model capable of identifying Twitter bots with high accuracy, and a web application for summarizing cybersecurity information from public data.

关键词： Computer engineering   Computer science  

摘要： As DRAM technology continues to evolve towards smaller feature sizes and increased densities, faults in DRAM subsystem are becoming more severe. Current servers mostly use CHIPKILL based schemes to tolerate up-to one/two symbol errors per DRAM beat. Such schemes may not detect multi-symbol errors arising due to faults in multiple data buses and/or chips. In this work, we introduce Single Symbol Correction Multiple Symbol Detection (SSCMSD) - a novel error handling scheme to correct single-symbol errors and detect multi-symbol errors. Our scheme makes use of a hash in combination with Error Correcting Code (ECC) to avoid silent data corruptions (SDCs). SSCMSD also enhances the capability of detecting errors in address bits. We develop a novel scheme that deploys 32-bit CRC along with Reed-Solomon code to implement SSCMSD for a x4 based DDRx system. Simulation based experiments show that our scheme effectively prevents SDCs in the presence of multi-symbol errors (in data) as well as address bit errors only limited by the aliasing probability of the hash. Our novel design enabled us to achieve this without introducing additional READ latency. We need 19 chips per rank (storage overhead of 18.75 percent), 76 data bus-lines and additional hash-logic at the memory controller.

关键词： Privacy   Protocols   Authentication   Internet of Things   Computer science   Handover   Security   privacy   trust   protocols   IoT   M-IoT   survey and analysis  

摘要： With an enormous range of applications, the Internet of Things (IoT) has magnetized industries and academicians from everywhere. IoT facilitates operations through ubiquitous connectivity by providing Internet access to all the devices with computing capabilities. With the evolution of wireless infrastructure, the focus from simple IoT has been shifted to smart, connected and mobile IoT (M-IoT) devices and platforms, which can enable low-complexity, low-cost and efficient computing through sensors, machines, and even crowdsourcing. All these devices can be grouped under a common term of M-IoT. Even though the positive impact on applications has been tremendous, security, privacy and trust are still the major concerns for such networks and insufficient enforcement of these requirements introduces non-negligible threats to M-IoT devices and platforms. Thus, it is important to understand the range of solutions which are available for providing a secure, privacy-compliant, and trustworthy mechanism for M-IoT. There is no direct survey available, which focuses on security, privacy, trust, secure protocols, physical layer security and handover protections in M-IoT. This paper covers such requisites and presents comparisons of state-the-art solutions for IoT which are applicable to security, privacy, and trust in smart and connected M-IoT networks. Apart from these, various challenges, applications, advantages, technologies, standards, open issues, and roadmap for security, privacy and trust are also discussed in this paper.

关键词： Adaptation models   Task analysis   Training   Decoding   Training data   Data models   Computer science   Neural network   dialogue system   domain adaptation   encoder-decoder model  

摘要： End-to-end task-oriented dialog systems have attracted vast amounts of attention in recent years, mainly because of their ease of training. However, such an end-to-end model requires a large number of labeled dialogs to train. Labeled dialogs are always difficult to obtain in real-world settings. We propose a domain adaptive end-to-end task-oriented dialog model that transfers knowledge in source domains to a target domain with limited training samples. Specifically, we design a domain adaptive filter in the encoder-decoder model to reduce useless features in source domains and preserve common features. A domain adaptive amplifier is designed to enhance the target domain impact. We evaluate our method on both synthetic dialog and human-human dialog datasets and achieve state-of-the-art results.

关键词： Computer science   Machine learning   Correlation (Statistics)--Data processing   Mathematical optimization--Computer programs   Computer algorithms  

摘要： Unsupervised representation learning algorithms have been playing important roles in machine learning and related fields. However, due to optimization intractability or lack of consideration in given data correlation structures, some unsupervised representation learning algorithms still cannot well discover the inherent features from the data, under certain circumstances. This thesis extends these algorithms, and improves over the above issues by taking data correlations into consideration. We study three different aspects of improvements on unsupervised representation learning algorithms by utilizing correlation information, via the following three tasks respectively: 1. Using estimated correlations between data points to provide smart optimization initializations, for multi-way matching (Chapter 2). In this work, we define a correlation score between pairs of data points as metrics for correlations, and initialize all the permutation matrices along a maximum spanning tree of the undirected graph with these metrics as the weights. 2. Faster optimization by utilizing the correlations in the observations, for variational inference (Chapter 3). We construct a positive definite matrix from the negative Hessian of the log-likelihood part of the objective that can capture the influence of the observation correlations on the parameter vector. We then use the inverse of this matrix to rescale the gradient. 3. Utilizing additional side-information on data correlation structures to explicitly learn correlations between data points, for extensions of Variational Auto-Encoders (VAEs) (Chapters 4 and 5). Consider the case where we know a correlation graph G of the data points. Instead of placing an i.i.d. prior as in the most common setting, we adopt correlated priors and/or correlated variational distributions on the latent variables through utilizing the graph G. Empirical results on these tasks show the success of the proposed methods in improving the performances of unsuper

关键词： Cognitive psychology   Artificial intelligence   Computer science  

摘要： This research presents a structure based on tabletop roleplaying for creating better computational models for simulating active-shooter events. Active-shooter events involve one or more people actively trying to kill others inside a populated and confined area. Roleplaying is the act of human participants portraying characters in a simulated environment. Roleplaying can be used for serious purposes, such as simulating real world events. Tabletop roleplaying adds maps and figures to show direction and relative positions of individuals and objects. These types of simulations are particularly beneficial when the real events occur infrequently or would be dangerous or costly to simulate with other methods. Versions of these simulations can include attributes for the human characters portrayed in the simulation. Attributes are quantified variables to distinguish differences between individuals, such as differences in size or athletic ability. These attributes can also include cognitive abilities and personality traits. A tabletop roleplaying structure was designed based on knowledge from multiple scientific fields. This structure determines the options and outcomes of actions attempted by characters in the simulated world. Data generated by this tabletop roleplaying structure are closer to data from actual active-shooter events than data from current AI (artificial intelligence) computational models. This is further improved when scientifically based attributes are incorporated. This tabletop roleplaying structure is defined in formulaic ways with quantified variables to be a template for creating more accurate computational models designed to simulate active-shooter events.

摘要： This thesis documents the development the Computer Science Academic Vocabulary List (CSAVL), a pedagogical tool intended for use by English-for-specific-purpose educators and material developers. A 3.5-million-word corpus of academic computer science textbooks and journal articles was developed in order to produce the CSAVL. This study draws on the improved methodologies used in the creation of recent lemma-based word lists such as the Academic Vocabulary List (AVL) (Gardner & Davies, 2014) and the Medical Academic Vocabulary List (MAVL) (Lei & Liu, 2016), which take into account the discipline-specific meanings of academic vocabulary. The CSAVL provides specific information for each entry, including part of speech and CS-specific meanings in order to provide users with clues as to how each item is used within the context of academic CS. Based on the comparative analyses performed in this study, the CSAVL was found to be a more efficient tool for reaching an minimal level of academic CS reading comprehension than the widely-used Academic Word List (AWL) (Coxhead, 2000), or the combination of the AWL with the Computer Science Word List (CSWL) (Minshall, 2013). Through coverage tests performed on a variety of corpora, CSAVL was shown to be representative of the written language of academic computer science and focused on the lemmas that are the most relevant to the context of written academic CS.

关键词： Computer science  

摘要： Most current authentication mechanisms rely on static initial verification of the user; however, such authentication mechanisms do not verify user identities on already unlocked systems. This thesis presents a continuous authentication mechanism that constantly examines user’s keystroke timing dynamics to assess the user’s identity. This mechanism preserves the privacy of the user by only utilizing the key press timings without storing information regarding which keys were pressed. It implements two one-class classifiers, only utilizing genuine user samples for training the model, and thereby mitigates adversarial inclusion of impostor data in the model. The data utilized in this pilot study consists of timing information from 20 users. One-class classification was performed independently for all the *** results show that in less than 85 characters, the performance results include a False Acceptance Rate (FAR) as low as 0.6% and a False Rejection Rate (FRR) as low as 1%.The overall implication of this work is that a slightly complex model is required to achieve good performance results when excluding the keys pressed by the user for classification.