教学课程资源库 更多>>

关键词： Artificial intelligence   Computer science  

摘要： Local learning methods are a popular class of machine learning algorithms. The basic idea for the entire cadre is to choose some non-local model family, to train many of them on small sections of neighboring data, and then to `stitch' the resulting models together in some way. Due to the limits of constraining a training dataset to a small neighborhood, research on locally-learned models has largely been restricted to simple model families. Also, since simple model families have no complex structure by design, this has limited use of the individual local models to predictive tasks. Recently, methods that do not directly employ local models, but that extract local information from data have proven to be very powerful in a number of tasks. Unfortunately, there has been little research on extending the local modeling schema to more complex model families and non-predictive tasks may prove valuable in many contexts. We hypothesize that, using a sufficiently complex local model family, various properties of the individual local models, such as their learned parameters, can be used as features for further learning. This dissertation improves upon the current state of research and works toward establishing this hypothesis by investigating algorithms for localization of more complex model families and by studying their applications beyond mere predictions. We summarize this generic technique of using local models as a feature extraction step with the term "local model feature transformations.'' The basic idea is that models that are trained on a local neighborhood of data 'encode' some information about the local structure of that neighborhood. This information can be extracted and used as features for input to other machine learning algorithms in a transfer learning scheme. The general intuition outlined above gives rise to a recurring prediction of our hypothesis that we will test in many experiments throughout this document. Namely, that existing machine learning methods

关键词： Mathematics   Theoretical mathematics   Computer science  

摘要： Hat guessing games—logic puzzles where a group of players must try to guess the color of their own hat—have been a fun party game for decades but have become of academic interest to mathematicians and computer scientists in the past 20 years. In 2006, Søren Riis, a computer scientist, introduced a new variant of the hat guessing game as well as an associated graph invariant, the guessing number, that has applications to network coding and circuit complexity. In this thesis, to better understand the nature of the guessing number of undirected graphs we apply the concept of saturation to guessing numbers and investigate the extremal and saturation numbers of guessing numbers. We define and determine the extremal number in terms of edges for the guessing number by using the previously established bound of the guessing number by the chromatic number of the complement. We also use the concept of graph entropy, also developed by Søren Riis, to find a constant bound on the saturation number of the guessing number.

关键词： Digital transformation   adaptive learning   computational thinking   computer science   education   information technologies   primary school   secondary school  

摘要： Schools, universities, and other educational entities are increasingly aware of the untapped potential of digital transformation, an essential process for increasing efficiency and collaboration, and reducing costs and errors in the management of at-scale training systems. In this context, the "Riconnessioni'' project was promoted by the Compagnia di San Paolo in agreement with the Ministry of Education but planned, started and developed by the Foundation for the School. The digital transformation started with a defined strategy that leveraged opportunities presented by new technology while meeting the objectives of system stakeholders. Through several steps, that strategy was developed for education connecting everything to support tomorrow's digital world and creating strong strategic partnerships able to build an ecosystem connecting people, processes, and things into a powerful, secure, and smart communications network. This paper reports on the three-year Riconnessioni project, which is combining the energies of teachers, managers, administrative staff, students, among others, and experimenting with new learning models, taking advantage of opportunities that emerged from perceptions stemming from concerns and systemic issues. To date, more than 150 schools in Italy have been included in the project, together with 550 teachers selected to scale up the instructional process. Using a methodology called "cascade training'', the 550 selected teachers were able to spread the knowledge to more than 2,600 colleagues. The monitoring and evaluation activity performed in Riconnessioni aims at processing information on implementation and results, following three lines. First, it regularly evaluates project activities from a reporting standpoint. Second, it verifies the plan consistency against implementation achievements. Third, it identifies changes produced and focuses on teachers' and students' skills to evaluate the effects of the project. The assessment framework is a

关键词： Artificial intelligence   Computer science  

摘要： Machine learning plays a pivotal role in most state-of-the-art systems in many application research domains. With the rising of deep learning, massive labeled data become the solution of feature learning, which enables the model to learn automatically. Unfortunately, the trained deep learning model is hard to adapt to other datasets without fine-tuning, and the applicability of machine learning methods is limited by the amount of available labeled data. Therefore, the aim of this thesis is to alleviate the limitations of supervised learning by exploring algorithms to learn good internal representations, and invariant feature hierarchies from unlabelled data. Firstly, we extend the traditional dictionary learning and sparse coding algorithms onto hierarchical image representations in a principled way. To achieve dictionary atoms capture additional information from extended receptive fields and attain improved descriptive capacity, we present a two-pass multi-resolution cascade framework for dictionary learning and sparse coding. This cascade method allows collaborative reconstructions at different resolutions using only the same dimensional dictionary atoms. The jointly learned dictionary comprises atoms that adapt to the information available at the coarsest layer, where the support of atoms reaches a maximum range, and the residual images, where the supplementary details refine progressively a reconstruction objective. Our method generates flexible and accurate representations using only a small number of coefficients, and is efficient in computation. In the following work, we propose to incorporate the traditional self-expressiveness property into deep learning to explore better representation for subspace clustering. This architecture is built upon deep auto-encoders, which non-linearly map the input data into a latent space. Our key idea is to introduce a novel self-expressive layer between the encoder and the decoder to mimic the "self-expressiveness'' property

关键词： Computer architecture   Software   Standards   Security   Computer science   Best practices   Code analysis   distributed systems   enterprise architecture   enterprise systems   global governance   microservice  

摘要： Code analysis brings excellent benefits to software development, maintenance, and quality assurance. Various tools can uncover code defects or even software bugs in a range of seconds. For many projects and developers, the code analysis tools became essential in their daily routines. However, how can code analysis help in an enterprise environment? Enterprise software solutions grow in scale and complexity. These solutions no longer involve only plain objects and basic language constructs but operate with various components and mechanisms simplifying the development of such systems. Enterprise software vendors have adopted various development and design standards;however, there is a gap between what constructs the enterprise frameworks use and what current code analysis tools recognize. This manuscript aims to challenge the mainstream research directions of code analysis and motivate for a transition towards code analysis of enterprise systems with interesting problems and opportunities. In particular, this manuscript addresses selected enterprise problems apparent for monolithic and distributed enterprise solutions. It also considers challenges related to the recent architectural push towards a microservice architecture. Along with open-source proof-of-concept prototypes to some of the challenges, this manuscript elaborates on code analysis directions and their categorization. Furthermore, it suggests one possible perspective of the problem area using aspect-oriented programming.

关键词： Computer science  

摘要： The University of New Hampshire Chemistry Department has been organizing study groups for students called Peer Led Team Learning (PLTL) for years now. However, assigning hundreds of students into groups manually is a non-trivial task, and doesn't guarantee the best possible solution. Each student will have certain times that they can meet, and more specifically, certain times that they prefer to meet. Additionally, when assigning students to groups, the students' traits should be taken in to account (e.g. preventing having only one freshman in the group). In this thesis, we devise a mathematical formulation of this problem, and apply an integer linear program solver to it to optimize the results. This work was developed with frequent input and feedback from the users to ensure that it met their needs and built on their suggestions. We demonstrate that our method is sufficiently fast to be used for the current needs of the users, and can continue providing high quality results as the future needs of the department scales.

关键词： Engineering   Energy   Computer science  

摘要： The continued integration of new sensing and control devices provides great opportunities to enhance the functionality of protective relaying, and thus reliability of electric power systems. This dissertation focuses on the development of multiple model filtering (MMF) fault diagnosis for transmission grids and islanding detection for distribution grids. Fault diagnosis identifies all short-circuited and open-circuited transmission lines, and islanding detection informs each distributed energy source whether it must disconnect from a distribution network due to an event elsewhere. The main objectives are to dynamically model a power system for the purpose of diagnosis/detection, to fully understand the structural interactions between the power system and its sensor network which supports diagnosis/detection tasks, to design the diagnosis/detection filters that effectively enhance the power system reliability, and to evaluate the performance robustness of the diagnosis/detection filters under various uncertainties. Using the concept of partition, a full-scale MMF-based fault diagnosis scheme is developed and tested for a 68-bus system to determine faulty transmission lines, fault types, and types of protection misoperations. Advanced digital relays capable of computation and communications are used to help implement a hierarchical diagnosis scheme over the sensor-localized transmission network. Both balanced and unbalanced faults are handled in time-domain because the more traditional phasor-domain analysis has become less reliable in modeling in the faster dynamics of today's power systems. Performance robustness is assessed with respect to fault location uncertainty. Computational complexity of the diagnosis filters is shown significantly reduced through network partition. The multiple model filtering concept is then applied to detecting the occurrence of unintentional islanding in distribution grids. The effectiveness of MMF-based islanding detection is illustrate

关键词： Information technology   Commerce-Business   Computer science  

摘要： The use of electronic/mobile-commerce (e/m-commerce) has become widespread and even fundamental to the global economy. However, users continued usage of e/m-commerce is not a guarantee if the infrastructural need to support e/m-commercial activities is not sufficiently provided. Quantitative methodology was used in the study. The purpose of this quasi-experimental study was to assess the relationships between e/m-commerce site access speed as provided by mobile network service providers (MNSPs) and new consumer adoption rates, as well as with existing consumer retention. The problem being studied in this study affects all e/m-commerce users as well as e/m-commerce companies. The variation in speed comes from simply changing from 2G speed through 3G speed to the 4G/LTE. Adoption/retention rates are measured based on a series of questions on the questionnaire and surveys. To analyze the results, different forms of parametric tests if assumptions are met otherwise non-parametric statistical tools were employed to determine significant relationships between the variables. Research data derives from interviews, questionnaires, and observations. Interviews, questionnaires, and observations would be administered on a large number of respondents such as e-commerce service providers, e/m-commerce users and intending users, to achieve at least a power of 0.78. The data analysis done in the study emanated from the data collected from 148 surveys completed by e/m-commerce users. From these data, the respondents were delineated into three groups. These include the Experienced-users, Averagely-Experienced users and non-experienced users. In the end, the findings of the study among others, primarily showed that mobile access speed is an important determiner of user adoption and user retention of e/m-commerce services.

关键词： Bilingual education   Technical communication   Science education  

摘要： Latinos are not engaging sufficiently in STEM careers, especially in science. Research studies on bilingual Latino students’ (BLS) learning in science suggest that educators’ expectations for Latinos to meet or exceed language proficiency and academic achievement standards are low. Data reported from the National Assessment of Educational Progress (NAEP) shows that new instructional methods, extra time, and strategies to pass high stake tests were not adequate to close the Latino achievement gap. Regardless of the persistent body of literature identifying the characteristics of effective schools, the BLS achievement gap continues. Latino school failure has been documented since the 1960s. Reasons for this situation include language and cultural differences; however, research two decades later demonstrated these were not the only unidimensional explanations facing Latino students’ educational failure. Instead, the situation is more complex and includes such circumstances as multiple social, political, and educational forces at work in schools. Nonetheless, research indicates bilingual children have a particular higher process of acquiring knowledge and understanding through their linguistic processing system, which allows for more than just linguistics proficiencies. But, the majority of bilingual Latino achievement gap studies have never been done in Puerto Rico, where bilingual schools are well established. More studies in Puerto Rico could provide a more suitable way to identify if the academic gap is due to language issues that persist among BLS in U.S. schools. Latinos are not engaging in STEM (science, technology, engineering, and mathematics) careers, especially in science. The purpose of this exploratory study was to implement a constructivism approach to teach an abstract science concept (i.e., velocity) using an innovative mini-lesson in two languages (Spanish and English) and a computer-based simulation (CBS), which serves as a manipulation in assessing th

关键词： Computer science  

摘要： Morphological changes of anatomical structures are widely studied as potential biomarkers that may allow researchers and clinicians to track disease progression, even before clinical diagnosis. This dissertation focuses on the development of novel methods to characterize morphological changes of anatomical structures in relation to biological or pathological processes, with emphasis on shape regression. Here, anatomical shapes are represented by the continuous medial representation (CM-Rep) rather than conventional surface representations. Regression analysis of CM-Reps leads to statistical analysis of rich geometric information, which could be hidden by extrinsic effects, decoupled from the morphological change, such as scale, local thickness, and bending. Shape features of CM-Reps do not only belong to a linear space so that standard regression analysis, which assumes the linearity of data, is not feasible for regression analysis of CM-Reps. The newly proposed regression methods, namely the geodesic regression model of CM-Rep, principal geodesic regression model, and hierarchical multi-geodesic model, account for the nonlinear characteristics of shape features by modeling them on the product of Riemannian manifolds. Applications to real-world data from a large clinical study for a Huntington’s disease, a neurodegenertative disease without a known therapy, demonstrate the ability and feasibility of the proposed methods for identification of potential biomarkers at the prodromal stage before clinical diagnosis. The proposed methods revealed significant group-wise differences between control and risk group subjects and also the effects of genetic risk and aging to longitudinal shape changes of the dorsal striatal complex, which is a crucial component of the motor and reward systems of humans. The proposed regression methods in this dissertation include both cross-sectional and longitudinal regression models, which are generic approaches with potentials to be applied