教学课程资源库 更多>>

关键词： Computer science   Artificial intelligence  

摘要： Poisoning attacks have been described as a grave danger to neural networks, they tend to change the complete meaning of the model and also define how the model interprets the data provided to it. Certain inputs are retrained after which poisoned data can be induced into the model. As the popularity of self-driving cars are growing exponentially, we need to look into every major flaw in the system and have security measures in place for the safety of the passengers. There have been limited studies on progressive poisoning attacks especially when it comes to autonomous driving vehicles. In this work, we first generate the poisoned data and also propose a generative method that will increase the generation of the poisoned data. This method of poisoning helps us determine the extent of inaccuracy and unexpected side effects when the poisoned data is fed to the model.

关键词： Computer science   Computer engineering   Web studies   Artificial intelligence   Information technology  

关键词： Computer science   Artificial intelligence  

摘要： Natural language generation (NLG) models' ability to generate long, fluent texts has enabled progress across many NLG subfields and increased the types of contributions models can make to human-machine collaborative writing tasks. However, the improved quality of generated text also poses challenges for NLG model evaluation. In this dissertation, we develop and evaluate methods for using NLG models in a collaborative setting to offer suggestions to people as they write. We identify new modeling directions for this setting and build one such model and demonstrate its effectiveness. Finally, we improve automatic and human evaluations for long, fluent generated text, both by developing and testing new automatic metrics and by evaluating the effectiveness of human evaluations for state-of-the-art language generation models. First, we explore the possibility of machine-in-the-loop creative writing. We performed two case studies using two system prototypes, one for short story writing and one for slogan writing. Participants found the process fun and helpful and could envision use cases for future systems. At the same time, machine suggestions do not necessarily lead to better written artifacts, and we suggest modeling and design choices that may better support collaborative writing. We explore one such direction (adding character representations as additional context for the model) and find it achieves improved generation results according to human and automatic metrics. We then consider the challenge of evaluating NLG models for collaborative writing and demonstrate how a collaborative writing platform can be used to collect pairwise, utterance-level human evaluations. For evaluating long machine-generated texts, automatic methods avoid the collection of human judgments, which can be expensive and time-consuming. We introduce methods based on sentence mover's similarity; our automatic metrics evaluate text in a continuous space using word and sentence embeddings. We fin

关键词： Neurosciences   Computer science   Bioinformatics   Medical imaging   Artificial intelligence  

关键词： Computer science   Artificial intelligence  

摘要： Clustering is a fundamental machine learning problem which seeks to discover groups of data based on a notion of similarity. Clustering is ill-posed as the notion of an optimal clustering is subjective to the application at hand. The clustering solution obtained depends on the characteristics of the data as well as on the design choices of algorithms. Clustering high-dimensional data poses additional challenges due to unreliable estimations of distances and density. High-dimensional data often reside in subspaces which entails discovering the dimensions relevant to each cluster. Another challenge in high-dimensional spaces is the emergence of the hubness phenomenon, whereby few data points, known as hubs, appear frequently as nearest neighbors. While certain hubs exhibit useful clustering properties of the data, others can negatively influence neighborhood computation and clustering results. As such, not all data points are beneficial for clustering, and more accurate and reliable clustering solutions can be obtained by leveraging an informative subset of data. In my dissertation, I propose data-driven and adaptive selection strategies to leverage exemplars and guide the optimization of unsupervised learning algorithms for data clustering. I introduce a new geometric characterization of hubs to guide the discovery of sub- space clusters, and introduce a hubness-driven algorithm to find subspace clusters in high-dimensional data. Furthermore, I leverage selective neighborhoods to approximate the data manifold and to regularize non-negative matrix factorization for data clustering. As a result, I design an unsupervised manifold regularized matrix factorization algorithm which jointly learns a sparse set of representatives and their neighbor affinities, along with the data factorization. I further propose a fast and effective approximation of my approach by relaxing the selectivity constraints on the data. Finally, data exemplars can be leveraged to learn unsupervised

关键词： Volatile organic compounds--VOCs   Genomes   Biomass   Metabolism   Lipids   Genes   Ecosystems   Greenhouse gases   Carbon   Glucose   Biomarkers   Algorithms   Metabolites   Ecosystem biology   Computer science   Genetics   Biogeochemistry  

摘要： Soil is the universal substrate which underpins agricultural productivity, providing plants and soil organisms with water and nutrient resources, as well as a plethora of additional anthropogenic ecosystem services. However, the sustained intensity at which we are using soil resources and the increasing frequency and intensity of extreme weather events is leading to a serious decline in soil quality, often defined as ‘the capacity of the soil to function’, and associated ecosystem function and service delivery. Better understanding and monitoring soil quality is key to slowing and reversing this decline. Soil biology (and related biochemistry) has often been underutilised as an indicator of soil quality; however, it is one of the most reactive and sensitive indicators. This thesis explores novel methods of profiling the small organic molecules (i.e., metabolites) in the soil; produced by the biological community during the catabolism of substrates and anabolism of cellular metabolites. It examines methods of profiling both primary (i.e., compounds involved directly in the growth, development and reproduction of organisms) and secondary (i.e., compounds performing additional functions)
metabolites in relation to soil quality and carbon (C) cycling. Specifically, I applied untargeted primary and secondary metabolomic methods to ‘real world’ field conditions and laboratory mesocosm experiments, assessing their applicability and aiming to further understand the complex biochemical interactions within the soil under a range of conditions, combining this data with a suite of physicochemical measurements to make conclusions about changes in soil quality and function. Here, I showed that, under field drought conditions, the primary metabolome shows similar trends to previous laboratory-based research, with significant increases in drought ‘biomarker’ compounds and storage lipids during drought, followed by a significant, rapid decrease in those compounds under post-drou

关键词： Computer science   Mathematics education   Educational technology   Instructional design  

摘要： With the increasing computer and network security challenges, Cyber Security Education is an important aspect to address the challenges. Game based learning when used in the curriculum engages the students but there are very few potential games that focus on advanced security topics for college students. In this thesis we developed an educational game to teach LAN vulnerabilities emphasizing on the Address Resolution Protocol. LAN vulnerabilities are stated as the top reason for insider threats and misuses of data by National Cyber Security and Communications Integration Center (NCCIC). It is critical for students to learn these vulnerabilities and know the common countermeasures. Therefore, we developed an interactive tool which would help the students to gain better understanding of the LAN Vulnerabilities and ARP concepts. The game comprises of three modules where the student will do an interesting role play and learn the concepts related to LAN vulnerabilities. The student will start with learning the basic LAN concepts and move towards learning about the Ethernet and how it is related to ARP, how ARP works and finally learn the counter measures for ARP. While learning the concepts, at the end of each concept, the student will be required to clear a quiz to go to the next learning module. After the completion of the game, the student will be completing a survey to capture their feedback on the Learning Experience and Objectives.

关键词： Eigen values   Algorithms   Dynamical systems   Artificial intelligence   Neural networks   Pattern recognition   Applied mathematics   Computer science  

摘要： Data is everywhere in the modern world. From a heartwarming grin to a bulk international commercial order, our lives can be easily reflected and captured by our activities on social media platforms, street security cameras, and internet footprints. As data becomes more accessible, our society inevitably turns to data-mining out of curiosity for self-exploration and value extraction for financial competition. However and as our data becomes both quantitatively and qualitatively complicated, methods that attempt to understand data are also changing. As of the time that this thesis is written, we are currently living in a world where Neural Networks are slowly becoming the new de-facto standard for understanding industry big data. Machine learning consists classes of modelling techniques that are capable of finding patterns directly from data itself. Of its many classes, neural networks have been gaining a large amount of following as they can learn in a near-automatic fashion with minimal amount of prior knowledge on the data. This property makes neural networks highly versatile and thus they are heavily favoured in becoming the new industry standard models. Despite their popularity, several neural network methods remain opaque and the research community still struggles to fully explain their behaviours. The lack of interpretability not only halts innovation in designing better models, but it also complicates the difficulty to automate machine learning (AutoML) for neural networks to work in a dynamic environment. To this end, this thesis is devoted towards increasing the understanding of neural network properties. More specifically, we will analyse neural networks through the mathematical lens of dynamical systems in order to clarify how they transform their input data. Then, the mathematical insights will assist us in addressing three interconnected topics – recurrent neural networks, meta-learning, and continual learning – to increase the flexibility and utility of

关键词： Computer engineering   Computer science   Artificial intelligence  

摘要： Data analytics has become the centre of novel research and extensively growing industrial applications. Data analytics contains various disciplines like statistical modelling, machine learning, deep learning, etc. Our research mainly focuses on the machine learning and deep learning components of the data analytics ecosystem. With the rapid growth of data, such data analytics workloads have focused more on the high-performance computing (HPC) paradigm. In general, the data engineering component holds the key to providing preprocessed data by operating on raw datasets in a data pipeline. Data engineering frameworks have also shifted towards high performance with the rapid growth of high-performance data analytical systems. Implementing data analytics-aware HPC data engineering operators is vital in providing scalable operators for HPC environments. This thesis focuses on data engineering beneficial for HPC environments. The critical factor is to provide a compatible software stack using HPC clusters efficiently. Mainly the HPC environments rely on communication via MPI. In addition to this, designing optimized compute kernels allows for the use of HPC resources effectively. This thesis considers three main areas: data engineering operators, interoperability, and usability. Data engineering operators discuss a set of widely used operators in data engineering. Interoperability focuses on the ability to be used by existing data analytics and data engineering systems, and usability details how HPC-aware data engineering operators are made available for efficient data exploration using the widely used dataframe abstraction. In data engineering, the most popular data abstraction is the dataframe. This thesis gives an in-depth analysis of a set of data engineering operators implemented to run on HPC resources. These operators are exposed to the user in terms of a state-of-the-art dataframe abstraction that involves less overhead in migrating an existing data engineering pro

关键词： Energy   Analytical chemistry   Chemistry   Computational chemistry   Computer science   Electromagnetics   Optics   Physics  

摘要： In this dissertation we analyze how Majorana quasiparticles found on material interfaces of both topological insulators (TIs) and topological superconductors (TSCs) are affected by imperfections within their local environment. While these quasiparticles are predicted to be critical for the construction of quantum computers, they are typically modeled only under pristine conditions. Thus, although quantum computers may require the spatial manipulation of Majorana quasiparticles, these topological material interfaces are commonly studied in static contexts and their response to manipulation remains an open question. We first demonstrate that interface potentials on the topological insulator Bi2Se3 can enable the emergence of Majorana bound states. Specifically, we study how non-helical spin-textures at the boundaries between TIs and SCs affect the proximity-induced superconductivity of the TI-interface state. We furthermore show that the Andreev conductance of lateral heterostructures joining TI-vacuum and TI-SC interfaces yields experimental signatures of the reduced symmetries of the interface. Next, we study the in-gap states that appear at the boundaries of both one-dimensional and two-dimensional TSCs. While massless Majorana quasiparticles are guaranteed to arise by the bulk-boundary correspondence, we find that they could be accompanied by massive Volkov-Pankratov (VP) states which are present only when the interface is sufficiently smooth. We calculate the spin-resolved local density of states of the VP states about the band inversion generated by a magnetic domain wall and find that they are oppositely spin polarized on either side of the topological phase boundary. Finally, we study how topological phase transitions in two-dimensional nanoflakes can be driven not only through magnetic domain walls but additionally through local changes in the system's chemical potential. We calculate the spatial extent of the one-dimensional Majorana modes circulating the bo