教学课程资源库 更多>>

关键词： Computer science   Information technology   Artificial intelligence  

关键词： Software packages   Materials science   Thermodynamics   Scale models   Grain growth   Carbon   Phase transitions   Statistical physics   Heat   Energy   Algorithms   Atoms & subatomic particles   Atomic physics   Computer science   Physics  

摘要： WC-Co cemented carbides combine superb hardness with high toughness making them ideal for usage in high-speed machining of steels and in wear resistance tools. These excellent mechanical properties are to a large extent dependent on the microstructure and thus the interfacial properties of the material. Hence, being able to predict and understand interfacial properties in this material can allow for, e.g., optimizing the manufacturing process in order to improve mechanical properties further. Electronic structure calculations allow for accurately predicting interface energies for a given structure and composition. However, finding the ground-state interfacial structure and composition is challenging as the search space is very large when considering all degrees of freedom. Furthermore, direct sampling of interfacial properties at finite temperature using density functional theory (DFT) is usually computationally impractical as hundreds, thousands or even millions of calculations may be required. Therefore, employing atomic-scale models based on DFT calculations is advantageous and allows for investigation of the interface structure, composition and free energy at finite temperatures. In this thesis computational methods for calculating temperature-dependent interfacial free energies are developed and applied to the WC-Co system. The emphasis is on understanding under which conditions cubic interfacial structures (complexions) can form on the WC basal plane in contact with Co. Configurational degrees of freedom are treated with cluster expansions and Monte Carlo simulations. Vibrational properties are mainly treated in the harmonic approximation using a regression approach to extract the harmonic force constants, which significantly reduces the number of DFT calculations. Interfacial phase diagrams are obtained for both the undoped WC-Co system and the Ti-doped system. Detailed information pertaining to structure and composition of the interfacial phases are obtained

关键词： Computer science   Artificial intelligence  

摘要： Software testing is an important step in the software development life cycle. It focuses on testing software functionalities, finding vulnerabilities, and assuring the software is executing as expected. Fuzzing is a software testing technique which feeds random input to programs and monitors for abnormal behaviors such as a program crash. Fuzzing can be automated and does not require access to the source code compared to manually reviewing the source code which requires a huge amount of time and cost. It can trigger vulnerabilities that the programmers overlooked while programming, such as buffer overflow, off by one error, etc. One of the limitations of fuzzing is that most programs require highly structured input or certain input patterns and therefore the fuzz testing may be terminated at an early stage of program execution because of not meeting the input format requirements. Some previous studies resolve this problem by manually creating program specific input grammars to help guide fuzzing, which is tedious, error prone, and time consuming. However, this solution cannot work efficiently when testing multiple programs which require different input patterns. To solve this problem, a general grammar-based fuzzing technique is proposed and developed in this dissertation. The new fuzzer can extract grammar from the sample input files of a program, and then generate effective fuzzing files based on the grammar. This fuzzing tool is able to work with different programs by extracting grammars from them automatically and hence generating program specific fuzzing files. The goal of this research includes developing an algorithm to extract grammars from sample input files, generating effective fuzzing files to test the programs, and implementing a fuzzing tool using Python programming language. The main contribution of this research is helping software developers and security experts in revealing vulnerabilities in various programs automatically by using the developed fu

关键词： Computer science   Artificial intelligence   Robotics  

摘要： The possibility of having artificial agents that can interact with humans and learn completely new tasks through instruction and demonstration is an exciting prospect. This is the goal of the emerging research area of Interactive Task Learning. Solving this problem requires integrating many capabilities across AI to create general robot learns that can operate in a variety of environments. One particular challenge is that the space of possible tasks is extremely large and varied. Developing approaches that cover this space is a difficult challenge, made more so by having to learn from a limited, though high-quality, number of examples given through interaction with a teacher. In this dissertation, we identify three major dimensions of task complexity (diverse types of actions, task formulations, and task modifiers), and describe extensions that demonstrate greater learning capabilities for each dimension than previous work. First, we extend the representations and learning mechanism for innate tasks so the agent can learn tasks that utilize many different types of actions beyond physical object manipulation, such as communication and mental operations. Second, we implement a novel goal-graph representation that supports both goal-based and procedural tasks. Thus the instructor can formulate a task as achieving a goal and let the agent use planning to execute it, or can formulate the task as executing a procedure, or sequence of steps, when it is not easy to define a goal. This also allows interesting cases of a task that blends elements of a procedure and goal. Third, we added support for learning subtasks with various modifying clauses, such as temporal constraints, conditions, or looping structures. Crucially, we show that the agent can learn and generalize a canonical version of a task and then combine it with these various modifiers within a task hierarchy without requiring additional instruction. This is done in the context of Rosie—an agent implemented within

关键词： Computer science   Artificial intelligence   Information Technology  

摘要： This research addresses the need for more innovation in cybersecurity, specifically the area of intrusion detection for distributed and complex operating environments (OE). Additionally, this research is a part of a multi-phased development for threat detection and risk mitigation to involuntarily execute an autonomic function for a device technology enabled network. The scope of this dissertation details the initial phase (Phase I) of the research, where a threat scenario within the device technology threat landscape is modeled. A security model for autonomic risk mitigation and threat detection of network constructs broadcasting the device technology data communications is presented, particularly devices classified as ‘things’. The security model is derived from a concept called neuroception, which is a subconscious method in the body for detecting threats. What is substantial about this method is that it provides a neural component to environment observation that uses the nervous system to discriminate whether people and situations are safe, dangerous or life threatening. The security model provides an enterprise platform cybersecurity architectural approach to detect and mitigate policy violations for the prescribed threat scenario within the Internet of Things (IoT) threat landscape. The operating environment (OE) of focus within this landscape is for the network-enabled device technology or ‘things’ propagating throughout a non-commercial, residential space or ‘smart home’. The security model proposes an intrusion detection security scheme that can involuntarily assist the given IoT network constructed in this phase (Phase I) to self-monitor the IoT network in future phases of this research, thus promoting an autonomic function. Providing distributed complex computing environments with an artificial neural capacity to environment sensation, can assist networks to self-regulate when the occurrence of adverse events is encountered. This capability is crucial to

关键词： Computer science   Applied mathematics   Artificial intelligence  

摘要： Optimal transport (OT) is a framework that can be used to facilitate the optimal allocation of resources in a network with multiple source and target nodes. To ease the computational complexity encountered by large-scale networks with a massive number of nodes, a distributed algorithm, based on the alternating direction method of multipliers (ADMM), is developed for computing the optimal transport strategy. However such a formulation lacks fairness, robustness and privacy considerations. Thus, there is an imperative need to develop distributed OT algorithms that allow for a more fair allocation of resources, accounts for possible deception attacks to the transport nodes and keeps nodes' sensitive information private during transport strategy updates. To achieve this goal, this thesis first incorporates a fairness metric into the objective function of the discrete OT problem and then leverages ADMM to develop a distributed algorithm. It then establishes a game-theoretic approach to counteract a deception attack where an attacker aims to compromise the transport plan. This formulation results in a min-max problem, and it can be solved in a distributed fashion to obtain a secure and resilient transport scheme. The distributed algorithms formed require communications on strategies between nodes during updates, which could potentially be intercepted and leveraged by an adversary, leading to private information being leaked. By incorporating differential privacy, the developed distributed algorithm guarantees the privacy of the sensitive information at each source and target node. All of the proposed algorithms are corroborated through case studies. The developed algorithmic foundation for fair, secure, and privacy-preserving discrete OT has broad applications to economics, machine learning and more.

关键词： Engineering   Computer science   Artificial intelligence   Energy   Neurosciences   Electrical engineering  

摘要： As computation increasingly moves from the cloud to the source of data collection, there is a growing demand for specialized machine learning algorithms that can perform learning and inference at the edge in energy and resource-constrained environments. In this regard, we can take inspiration from small biological systems like insect brains that exhibit high energy-efficiency within a small form-factor, and show superior cognitive performance using fewer, coarser neural operations (action potentials or spikes) than the high-precision floating-point operations used in deep learning platforms. Attempts at bridging this gap using neuromorphic hardware has produced silicon brains that are orders of magnitude inefficient in energy dissipation as well as performance. This is because neuromorphic machine learning (ML) algorithms are traditionally built bottom-up, starting with neuron models that mimic the response of biological neurons and connecting them together to form a network. Neural responses and weight parameters are therefore not optimized w.r.t. any system objective, and it is not evident how individual spikes and the associated population dynamics are related to a network objective. On the other hand, conventional ML algorithms follow a top-down synthesis approach, starting from a system objective (that usually only models task efficiency), and reducing the problem to the model of a non-spiking neuron with non-local updates and little or no control over the population dynamics. I propose that a reconciliation of the two approaches may be key to designing scalable spiking neural networks that optimize for both energy and task efficiency under realistic physical constraints, while enabling spike-based encoding and learning based on local updates in an energy-based framework like traditional ML models. To this end, I first present a neuron model implementing a mapping based on polynomial growth transforms, which allows for independent control over spike forms and t

关键词： Computer science   Neurosciences   Artificial intelligence   Datasets   Electroencephalography   Neural networks   Signal processing  

摘要： Motor imagery (MI) has been one of the most used paradigms for building brain-computer interfaces (BCI), widely used in neurorehabilitation, for restoring functionality to damaged parts of a neurologically deficient person. The existing motor imagery techniques have largely employed feature extraction techniques such as the power spectral density (PSD) and the common spatial patterns (CSP) before classification, using traditional machine learning algorithms such as support vector machines (SVM) and linear discriminant analysis (LDA). These algorithms are quite limited in their ability to generate feature representations for certain types of signals, limiting the potential for improvements in the decoding process. Also, the problem of signal non-stationarity inherent in many neurological studies typically requires calibrating a classifier for each subject and session of the study, tantamount to longer wait times and possibly rendering the BCI impractical to use. Another challenge with current decoding techniques is the inadequacy in interpretations and verification of the learnings of the classifiers. This work focuses on using more recent deep learning-based techniques with data augmentation for the decoding process, transfer learning to handle the non-stationarity across subjects and sessions and interpretability, validating the model’s learnings. Using public motor imagery electroencephalography (EEG) datasets, we demonstrate that specific feature engineering is not needed with deep learning-based techniques and that transfer learning can be used to handle the intra- and inter-subject non-stationarity of signals. Furthermore, we provide interpretations using region contribution analysis, a framework based on occlusion sensitivity, to show that the model learnt task-relevant features.

关键词： Computer science   Mining  

摘要： Software developers are increasingly having conversations about software development via online chat services. Many of those chat communications contain valuable information, such as description of code snippets and APIs, opinions on good programming practices, and causes of common errors/exceptions. Researchers have demonstrated that various software engineering tasks (e.g., recommend mentors in software projects, aid API learning) can be supported by mining similar information from other developer communications such as email, bug reports and Q&A forums. However, limited work has focused on investigating the availability and mining of information from developer chats. To successfully mine developer chat communications, one has to address several challenges unique to chats. The nature of chat community content is transient. Developer chats are typically informal conversations, with rapid exchanges of messages between two or more developers, where several clarifying questions and answers are often communicated in short bursts. Chats thus contain shorter, quicker responses, often interleaved with non-information-providing messages. As a result, it is difficult to find relevant information in a large chat history, and important advice is lost over time. My thesis is that software developers communications through online chat forums are a valuable resource to mine and valuable knowledge can be automatically mined from this resource towards improving and building new tools to support software engineers. The focus of this dissertation is Mining Information from Developer Chats Towards Building Software Maintenance Tools: (1) As a first step towards mining, we investigated the availability of information in chats through an empirical study. We also analyzed characteristics of chat conversations that might inhibit accurate automated analysis. (2) Next, we extended an existing algorithm to automatically extract (or disentangle) conversations for analysis by researchers or a

关键词： Physics   Medical imaging   Computer science   Fluid mechanics   Mathematics   Mechanics  

摘要： Foam is a dispersion of gas in a liquid, where the liquid is continuous and part of the gas is made discontinuous by thin liquid films (lamellae) (Hirasaki and Lawson, 1985). Foam promises to substantially improve reservoir sweep in gas-injection enhanced oil recovery (EOR) projects. Predictive modeling efforts are hindered by many complex rheological characteristics shared by foams, including a yield stress, non-Newtonian viscosity, finite shear modulus and slip at the wall, etc. (Reinelt and Kraynik, 1989).This dissertation focuses on experimental and numerical simulations on foam physics across multiple scales. The first objective is to develop ultimately a mechanistic, physical-based pore network model to improve the understanding of multiple pore-level mechanisms involved during foam propagation in porous media. To this aim, a pore network model is developed to simulate foam propagation as a drainage process of gas invasion into a porous media initially saturated with a surfactant solution. During this process, lamella generation, destruction, and mobilization are involved. A modified invasion percolation with memory (MIPM) is proposed to take the effects of lamella destruction into account. Specially, lamella generation is incorporated by snap-off, leave-behind, and division mechanisms. Lamella destruction by capillary suction at the plateau borders is modeled using the Reynolds equation for film thinning and lamella rupture is assumed to occur when the film thickness falls below a certain critical thickness at which the maximum disjoining pressure is attained. This mechanism is incorporated in the pore network model to which we add a notional time-dependency of the invasion percolation with memory *** the onset of foam mobilization, the drag on moving lamellae imparts an additional shear-thinning contribution to the effective gas viscosity. To take viscous flow effects, which arise from the drag on moving lamellae following the onset of foam mobil