教学课程资源库 更多>>

关键词： 疫情   初中英语   线上教学   未来职业规划  

摘要： 面对2020年初突如其来的新冠疫情,各行各业的人都在用自己的方式为祖国加油助力。作为一线初中英语教育工作者也不能缺席。走出传统课堂教学,进入以"网络虚拟空间"为课堂载体的线上教学,走上荧幕,当上"主播",运用网络课程资源平台进行线上教学。将抗击疫情与英语学科教育相结合,通过客观事实的呈现,让学生了解各行各业对社会发展的重要影响,引导青少年正确认识"职业",关注科学技术的重要性。

关键词： Bioinformatics   Genetics   Computer science  

摘要： Gene expression regulation is a set of critical biological processes that give rise to the diversity of cell types across tissues and development stages. Noncoding regions of the genome (intergenic + intronic, > 98% of genome) play an important role in these processes, with noncoding genetic variation quantitatively affecting transcriptional activity, splicing of pre-mRNA, and localization, stability, and translational control of mRNA transcripts. Previous genetic studies of human disease have implicated numerous common noncoding loci with small but significant effect in common conditions. Recently, we and others have reported evidence supporting a role of rare noncoding variants with larger effect in early onset conditions such as birth defects and neurodevelopmental disorders. These early onset conditions are quite common in aggregate, affecting over 3% of young children. A better understanding of the functional impact of rare regulatory noncoding variants will enable novel genetic discovery, give insights of disease mechanisms, and ultimately improve diagnosis, treatment, and clinical care. In this thesis dissertation, I describe three related projects. First, we used a combinatorial multi-testing framework to find excess burden of noncoding de novo mutations in congenital heart disease (impacting both transcriptional and post-transcriptional regulatory stages). This finding was central to the rest of my work, motivating the development of new computational approaches to predict genetic effect of noncoding variants through the lens of post-transcriptional regulation. Second, we used convolutional neural networks to model and understand sequence specific RBP binding processes. Finally, we designed a graphical neural network model capable of integrating cause and consequence to predict genetic effect of rare noncoding variants. In summary, we developed new machine learning methods to analyze multimodal human genome sequencing data, uncover deeper insights into post

关键词： Computer science   Computer engineering   Artificial intelligence  

摘要： In the field of deep learning, attention mechanism is an approach that imitates human cognition. The mechanism helps a model to focus on important portions of input, with the intuition that allocating more resources towards important portions of data improves model performance. It is widely used to model sequential data in natural language processing. An attention model creates short cut linkages between distant positions in a sequence. The linkages are dynamically learnt in an end to end manner. It has been proved effective for context information extraction in text. Attention models are empirically formulated previously based on heuristic design of network architecture. This work provides analytical formulation on the model to gain knowledge on their functioning dynamics, convergence speed and generalization ability. We propose to analytically formulate attention models with attention neural tangent kernels (ANTK). ANTK characterize the behavior of deep neural networks based on kernel methods, which utilize kernel functions to learn proximity information in data points. It enables tractable solutions on network predictions, training dynamics, generalization etc. The neural tangent kernel is formulated as Gaussian process to produce a recursive derivation of its expression for each layer of the network. This leads to a new model built on kernel methods that approximates the behavior of attention model. Dynamics of the neural model can be approximated (with proof of convergence) by analyzing the formed kernel without performing gradient descent. We evaluate ANTK on aspect based sentiment analysis, which is a challenging task in natural language processing. The goal of sentiment analysis is to interpret and classify emotions expressed in text data. It plays an important role in applications such as product recommendation, public opinion detection and human machine interaction etc. Aspect based sentiment analysis further breaks down text into different aspects (attrib

关键词： Computer science   Computer engineering   Artificial intelligence  

摘要： As modern GPU workloads become larger and more complex, there is an ever-increasing demand for GPU computational power. Traditionally, GPUs have lacked generalized data-dependent parallelism and synchronization. In recent years, there have been attempts to introduce a more sophisticated form of synchronization between different kernels in an application to control the flow and ensure the correctness of the outputs. However, coarse synchronization between such kernels can significantly reduce GPU utilization. Moreover, with hundreds or thousands of kernels in a workload, the overhead can be consequential. Due to GPU’s massive parallel design, data can be split among thread blocks, which allows us to manage the data dependencies on a more fine-grained level between the thread blocks themselves rather than the kernel containing them. In this dissertation, we propose several methods to improve the performance of data-dependent GPU applications in this fashion. In our first method, Wireframe, we propose a hardware-software solution that enables generalized support for data-dependent parallelism and synchronization. It allows dependencies between the thread blocks in the GPU kernel to be expressed through a global dependency graph, which is then sent by the GPU hardware at kernel launch, which then enforces the dependencies in the graph through a dependency-aware thread block scheduler. Our second method, BlockMaestro, is aimed at improving the user transparency in the process of determining the inter-kernel thread block dependencies through static analysis of memory access patterns at kernel-launch time. During the runtime, BlockMaestro enables kernel launch hiding by launching multiple kernels on the GPU and utilizes a thread block scheduler in hardware to schedule the thread blocks with satisfied dependencies for execution. In our third method, SEER, we aim to expand our support for data-dependent applications to those with non-static memory accesses, which can only be

关键词： Geography   Computer science   African studies  

摘要： Despite the growing use of electronic products, waste recycling efforts are not expanding at a proportional pace, and places in the Global South like Ghana are left with toxic waste sites like the one at Agbogbloshie that are arguably an indirect result of Western consumer capitalism. In spite of significant international attention, the e-waste trade in Agbogbloshie continues to persist, prompting the need for a more comprehensive look at the potential impacts of its continuation on the environment and human health. By examining the potential impacts of the e-waste trade in Agbogbloshie through a critical geography lens, it is hoped that a more nuanced account can assist restoration efforts that minimize harm to the communities that live in and rely on the e-waste trade. This thesis presents an analysis of seven video interviews specific to different careers concerning e-waste in Agbogbloshie with a particular emphasis on the identification of the potential impacts of the e-waste trade. Utilizing transcribed interview footage alongside existing literature and applying qualitative analysis techniques, several conclusions were reached based on the main thematic points identified: workers know e-waste is harmful to their health, e-waste is a significant source of income for many people, the disconnect between the formal and informal sectors makes restoration efforts difficult to coordinate, and the future of skilled workers is in jeopardy. Although far from comprehensive given the small sample size, these interviews lend a great deal of insight into the potential human and environmental impacts of electronic waste in Agbogbloshie.

关键词： Computer engineering   Computer science   Artificial intelligence  

摘要： he ending of Moore’s law and Dennard scaling (MOSFET scaling) law has shifted the focus of academia and industry towards hardware accelerators and domain-specific architectures (DSA). Specialized hardware components are increasingly infused at the various computing levels. In particular, graphics processor units (GPUs) are now considered the most common hardware used to accelerate scientific computations at all scales, from embedded to exascale systems. The increasing computational demands in areas like high-performance computing (HPC), machine learning (ML) and specifically deep learning (DL), computational molecular dynamics (CFD), and big data analytics have pushed GPUs over the top in popularity and versatility way beyond gaming. The broader use of GPUs has motivated the research community to invest in developing modeling and simulation (ModSim) tools. Such tools are critical in exploring the design space of the hardware. It also guides architects and developers in optimizing their applications’ performance by doing performance characterizations and sensitivity analysis on the workloads. However, with complexities found in the microarchitecture of various GPU architectures, having a fast and accurate ModSim is a non-trivial task. The currently available models represent good progress. Nevertheless, they are either limited to a specific GPU generation, very slow, focus on a particular hardware component, or lack prediction accuracy. A verified GPU performance model is still an open problem in terms of accuracy, comprehensiveness, and scalability. This dissertation solves the limitations in the current ModSim tools by presenting PPT-GPU, Performance Prediction Toolkit for GPUs. PPT-GPU is a comprehensive end-to-end modeling tool that predicts the performance of contemporary GPU architectures without scarifying the accuracy of prediction or the speed of simulation. PPT-GPU achieves scalability through a hybrid high-level modeling approach where some computations ar

关键词： Computer science   Public health   Computer engineering   Biomedical engineering   Bioinformatics  

摘要： Technological advancement remains on the increase in every sector in the business sphere globally. The same can be said about the healthcare sector where data distribution and management among other operational areas has been improved via technological advancement. While these improvements have brought positive changes in the healthcare industry, some challenges remain unsolved. The national ecosystems of the healthcare industry do not have a common system where critical patient data can be securely shared and be made available for use among service providers or medical silos, irrespective of their location. One of the capable technologies, though young, but has proven matured enough to handle secure data exchange is Blockchain. This study researched the strategies for implementing blockchain technology for a secure integrated data exchange system that facilitates healthcare providers to share patients’ medical data among their peers over a decentralized network nationally. After successful identification and execution of the strategic integrated data exchange implementation via blockchain, patient medical data could be readily available at any medical silo of the patient’s choice during their visit. Besides, this integrated data exchange system implementation could mitigate the dispersed patient data within the healthcare ecosystems across the nation, and the patient will not have to constantly provide their medical history verbally or via form-filling during their visit to especially new healthcare providers. In the nutshell, this process could accelerate patient care delivery and could also enhance diagnostic accuracy, furnished by the patient’s medical history, treatments, and prescribed medications by previous healthcare providers.

关键词： Semantics   Quality of service   Web services   Search engines   Testing   Computer science   Oceans   Acceptance testing   interface compatibility   RESTful service   service discovery   term expansion   unit testing  

摘要： Web API/service technology has been attracting considerable attention and the REST (REpresentational State Transfer) architecture is now widely accepted as mainstream technology. Nonetheless, developing the means by which to discover RESTful Web APIs/services is crucial to the further development of this technology. Unfortunately, existing search engines for RESTful Web APIs/services provide only keyword-based or tag-based search functions. A failure to take into account the semantics and/or characteristics (e.g., their interface compatibility) greatly hampers the ability to find suitable APIs/services. In this study, we propose a novel approach to the discovery of RESTful Web API/services, referred to as Test-Oriented API Search with Semantic Interface Compatibility (TASSIC). This scheme involves expanding the terms of Web API/service documents based on DBpedia and WordNet. Unsuitable APIs/services are then filtered out using a systematic process, as follows: 1) calculation of semantic similarity between a set of candidate APIs/services and a user query, 2) analysis of interface compatibility between candidate APIs/services using the Hungarian Algorithm, 3) invocation of candidate APIs/services to verify functionality and availability, and 4) analysis of similarity between the actual response of the candidate services and the expected response specified in the user query. The proposed TASSIC increases the likelihood of matching APIs/services that are semantically equivalent or similar to user queries. Unit test and acceptance test are used to verify that a set of candidate APIs/services are actually available and that they actually meets user requirements. Experiment results demonstrate the efficacy of TASSIC, the accuracy of which is superior to that of existing methods.

关键词： Computer science   Artificial intelligence   Electrical engineering  

摘要： Algorithms are used to make decisions in an ever-increasing number of socially consequential domains. From risk assessment tools in the criminal justice system to content moderation tools to assessments in hiring, algorithms play a key role in shaping the lives of people around the world. Algorithms offer many potential benefits: they are consistent, scalable, and can leverage more data than any human could reasonably consume. However, without careful consideration algorithmic decision-making also carries a number of risks, like replicating human biases, creating perverse incentives, and propagating misinformation. This thesis seeks to develop principles for the responsible deployment of algorithms in applications of societal concern, realizing their benefits while addressing their potential harms. What does it mean to make decisions fairly? How do theoretical ideas about societal impacts manifest in practice? How do existing legal protections apply in algorithmic settings, and how can technical insights inform policy? In this thesis, we explore these questions from a variety of perspectives. Part II leverages theoretical models to surface challenges posed by algorithmic decision-making and potential avenues to overcome them. Part III incorporates models of behavior to better understand the interplay between algorithms and humans decisions. In Part IV, we explore how these insights manifest in practice, studying applications in employment and credit scoring contexts. We conclude in Part V with open directions for future research.

关键词： Design   Computer science   Electrical engineering   Energy   Engineering