一种语言模型的精确学习方法

语言模型具有很好的可解释性，但面对复杂问题时，其精确性不能满足要求。为此结合微粒群算法和遗传算法各自的演化特点，采用两阶段学习策略。对语言模型进行分层演化，首先利用微粒群算法优化各输入变量的语言值数目及对应的模糊集参数。形成候选规则集，再应用遗传算法选择规则，得到精确的语言模型．该方法几乎无需先验知识，可直接从样本数据获取语言模型，应用函数近似为例说明其有效性。     

From Teaching to Learning: Technological Potential and Sustainable, Supported Open Learning

The central theme of this paper is the current interest in most educational institutions in moving from teaching to learning as their main system model and the implications which technology media have for unravelling the debate and influencing the resulting practice (for example see Active Learning: Using the Internet for Teaching, Number 2, July, 1995), Our chosen strategy for dealing with the central theme is to consider how we use language, metaphor and models to describe systems for teaching and learning and what is the role of technology in these systems. In particular we describe how the Open University is moving from an analysis of individual technologies to a synthesis of the educational ideas into a sustainable system that conforms to the University's policy of supported open learning. This includes focusing on open and equal access to courses, considerable attention to staff development and training (specifically in core teaching areas such as face to face tuition, correspondence tuition, student support, telephone and other media supported communication) as well as the necessity and value of reflection on practice (e.g. see Baker, Tomlinson et al,, 1996). In this first section, taking as our starting point the traditional linear view of educational structures, we build upon the notion of the learning system and describe this as an approach with a somewhat long and surprising history originating in the Socratic method. An eductive, cyclic learning model is introduced, and the historical impact of technology on this model is briefly reviewed in a global context by addressing the specific issue of access from the developing countries. Following from this, in sections two and three two types of teaching are described, under the labels of conventional and distance. It is argued that each has strengths and tendencies towards the eductive learning system introduced in section one. However, it is also argued that each has flaws which provide problems for the development of a learning system which can be sustained at distance. An analysis based upon metaphor is applied. In section four, it is then argued that in the conventional model there is a system with excellent learning potential but with an organisational structure which often does not encourage it whereas in the distance model there is the potential organisation for providing feedback but often little effective use made of it. That lack of use includes the new technologies which are now being focused on by the Open University, both in terms of the INSTILL initiative (Integrate New Systems and Technologies Into Lifelong Learning) and the Technology Strategy for Academic Advantage (Laurillard, Christmas et al., 1996). These technologies are argued, in section five, to be capable of being tactically effective in stimulating and supporting the strategic aim of learning. Technology mediation is discussed both in terms of course-based teaching and in terms of research opportunities. In section six a range of problem/opportunity areas of immediate concern for the extension of this technologically mediated system to the developing countries are discussed, while policy implications for the extension of supported open learning are drawn out in section seven. These include policies toward co-learning, access to learning, quality standards and the authorship of educational material.     

Learning to Predict in Complex Biological Domains

1 Introduction1 Bioinformatics is a rapidly expanding field involving a significant contribution from the data mining community. Many different problems are grouped into the general category of bioinformatics, including protein secondary structure prediction, drug discovery, DNA microarray analysis, gene prediction and genome analysis [17]. In this paper, we focus on two particular problems in which data mining can contribute: drug discovery and protein secondary structure prediction. Both …     

Enriching Service Learning by its Diversity: Combining University Service Learning and Corporate Social Responsibility to Help the NGOs Adapt Technology to Their Needs

Community-based service learning helps students link subject matter to everyday life and developing sense of responsibility to their community. Corporate social responsibility (CSR) is a form of corporate self-regulation integrated into a business model. Both promote civic engagement but the two hardly come across each other. We explore the process that enriched service learning with its diversity by combining CSR practices of a multinational software company. It begins by exploring different forms of action research with a focus on action engagement to improve students’ involvement in marginalized communities. The article provides field-based reflections of the sustained engagement and suggests ways in which service learning experiences may create rooms for innovation and growth in the non-profit sector, local communities, businesses, and students themselves.     

Practical Contributions of the Systems-Based Evolutionary Learning Laboratory to Knowledge and Stakeholder Management

This paper reports on a case study on improving the quality of life for women smallholder farmers in Vietnam, using the systems-based Evolutionary Learning Laboratory (ELLab) approach. The study aims to provide evidence of practical contributions of the ELLab to project stakeholder and knowledge management through the case study. A transformative learning framework was developed to provide initial reflections on the planning phase of the project. Practical contributions of the ELLab process to stakeholder identification and management, and knowledge creation during project planning and implementation are discussed. Strong evidence was found of transformations in awareness and perspectives amongst project actors that subsequently led to systemic and executable action plans. This clearly proved the value of the ELLab approach in informal transformative learning.     

提高BP网络学习速度的自适应算法

针对BP网络存在收敛速度慢及目标函数容易陷入局部极小值的缺点,本文首先对前人所做的改善BP网络学习速度的方法进行了探讨,其次分析了学习率对BP网络学习速度的影响,最后提出了一种提高BP网络学习速度的新方法,即自适应调整学习率和动量因子.计算结果表明,该方法大大地提高了收敛速度,而且算法简单、易行.     

Action Research to Enrich Learning in e-Tutoring for Remote Schools

Systemic Practice and Action Research - Because students in remote areas face unequal learning opportunities, the Ministry of Education (MOE) in Taiwan commenced a benchmark program in 2006 known...     

Using the ‘Engagement’ Model of Problem Solving to Assist Students in Capstone Learning

The engagement model for problem solving was developed as a response to soft systems methodology being difficult to implement in certain circumstances. In the years since the original paper was published, the model has developed into an iterative problem solving approach for the assessment, creation and implementation of new ideas in difficult problem situations. This case study paper exemplifies the kind of work undertaken through the use of the model and provides a practical scenario for how one might apply it. The case study takes place in a higher education environment and relates to the development of a capstone course for management majors. The paper concludes with suggestions for future research.     

最小二乘支持向量机在强化学习系统中的应用

将连续状态空间下的Q学习构建为最小二乘支持向量机的回归估计问题,利用最小二乘支持向量机良好的泛化以及非线性逼近性能实现由系统状态-动作对到Q值函数的映射。为了保证计算速度以及适应Q学习系统在线学习的需要,最小二乘支持向量机的训练样本是窗式移动的,即在Q学习系统学习的同时获取样本数据并进行最小二乘支持向量机的训练。小车爬山控制问题的仿真结果表明该方法学习效率高,能够有效解决强化学习系统连续状态空间的泛化问题。     

多维非线性时间序列的投影寻踪学习逼近

建立了多维非线性时间序列投影踪学习网络结构及算法,证明了投影寻踪学习网络可以以任意精度逼近多维非线性时间序列。解决了基于投影寻学习的多维非线性时间序列的建模和预测问题,实际应用例子表明该算法可行。     

基于性能预测的遗传强化学习动态调度方法

针对作业车间动态调度问题,在模式驱动调度的框架下,提出遗传强化学习动态调度方法。首先,采用优先规则编码的染色体表达问题的解,将染色体分割成基因模式作为分阶段调度算法的状态模式;其次,设计性能预测变量,构建启发式立即回报函数,引导和加快遗传强化学习算法的搜索进程;再次,设置遗传算子、强化学习及其相关参数以实现搜索过程＂开采＂与＂探索＂之间的平衡;最后,仿真实验结果验证了遗传强化学习调度方法的有效性。

Abstract：

In the framework of pattern driven scheduling,a genetic reinforcement learning （GRL） approach to schedule the job in the dynamical job-shop was proposed.First,the chromosome was coded by preference rules-based representation for the problem.The chromosome was divided into gene schema as state patterns for the multi-phase scheduling system.Secondly,a performance predictive variable to construct instant reward function was designed which was used to guide the learning system to progress rapidly.Thirdly,genetic operators,RL and controlling parameters carried out the search strategy for the balance of ＂exploration＂ and ＂exploitation＂.Finally,the simulation results verify the efficiency of GRL scheduling approach.     

基于扩展Kalman滤波的神经网络学习算法在股票预测中的应用

给出一种新颖的用于股价预测的基于扩展Kalman滤波的神经网络学习算法。与传统的BP算法相比，该方法具有更好的收敛率和学习能力。通过对股票的预测实验验证该方法的可行性和有效性。     

具有学习效应的总完工时间流水作业问题

研究工件具有学习效应的2台机器流水作业排序问题.工件的学习效应指工件的加工时间为所排位置的指数函数.目标函数为极小化总完工时间.给出该问题的数学规划模型.同时对大规模问题给出3个启发式算法,计算结果表明,用这3个算法解决所研究问题比较有效.     

基于模糊神经网络局部强化学习在Robocup中的应用

针对Robocup仿真组比赛中智能体的配合与动作选取,将模糊神经网络（FNN）和局部协调图动态角色分配与传统Q-学习相结合,提出了基于模糊神经网络的局部Q-学习。采用该方法,有效抑制了仿真平台中的噪声干扰,提高了动作选取的精度,解决了传统Q-学习中Q表占用内存空间过大的问题,增强了系统的泛化能力,并进一步缩短了学习时间,更好的满足比赛实时性的要求。将其运用于仿真组比赛的传球和射门模型中,验证了该方法的有效性。     

A Two-Dimensional Approach to Iterative Learning Control with Randomly Varying Trial Lengths

In this paper, iterative learning control(ILC) is considered to solve the tracking problem of time-varying linear stochastic systems with randomly varying trial lengths. Using the two-dimensional Kalman filtering technique, the authors can establish a recursive framework for designing the learning gain matrix along both time and iteration axes by optimizing the trace of input error covariance matrix.It is strictly proved that the input error converges to zero asymptotically in mean square sense and thus the tracking error covariance converges. The extensions to that prior distribution of nonuniform trial lengths is unknown are also investigated with an asymptotical estimation method. Numerical simulations are provided to verify the effectiveness of the proposed framework.     

Using Critical Social Heuristics and Project-Based Learning to Enhance Data Warehousing Education

The aim of this paper is to demonstrate how project-based learning can be used from a critical systems perspective in data warehousing education. Data warehousing is a discipline in information technology focusing on providing data-driven decision support systems for strategic decision making. In this study we used action research from a critical (emancipative) perspective to reflect on our current instructional design of the data warehousing module before redesigning it to better serve the needs of the involved and affected. We used critical systems heuristics and project-based learning as frameworks of understanding to guide our intervention. Project-based learning is a learning/teaching approach aimed at organising the learning experience in terms of a project. We used written interpretive interviews in the diagnosis and evaluation of success phases of our action research cycle. Our reflection is according to the action research model of Checkland reflecting on our success in the area of application (data warehousing instruction) as well as our methodology (action research from a critical social theory perspective) and our framework of ideas (project-based learning and critical social heuristics).     

匹配于进化种群的树形贝叶斯网络

为了构造匹配于进化种群的树形模型，首先研究了基于贝叶斯狄利克雷度量构建树形网络结构的方法，得出关键在于搜索每一个节点的最大值父节点。然后提出了节点的势及对称节点概念，证得节点的势与网络结构的连接方向关系密切，以及节点与其对称节点在贝叶斯网络图中具有相同的度量属性。最后给出了仿真分析结果，进一步表明本文提出的方法能够依据数据信息搜索到具有最大度量值的树形网络结构。     

基于多传感器数据融合的机动目标跟踪自适应学习方法

考虑到传感器对目标的观测都存在有偏差，同时针对雷达系统检测中存在漏检的现象，提出了一种适用于处理传感器漏检现象的多传感器融合方法。并以雷达／红外成像复合的双模制导体系下红外和雷达两种传感器对目标状态的检测和跟踪为例，给出一个完整的多传感器数据融合和目标跟踪自适应方法。该方法简单易行，鲁棒性强，通过对红外成像和雷达系统的数据融合和目标跟踪的数字仿真，表明该方法有效。另外，该方法同样适用于多于两个多传感器系统。     

基于案例学习和目标驱动学习的观点生成框架

群体观点生成是观点评价和群体一致性意见达成的基础．本文提出基于案例学习和目标驱动学习的观点生成框架,将人工智能技术集成入群体决策支持系统中,通过动态问题分解过程、多案例经验学习过程、主动的目标驱动学习过程的协调运作,既支持初始观点生成,又支持观点的修改与扩充．本文采用描述性智能辅助技术,使得观点生成的智能辅助方法不仅易实现,而且具有高度的灵活性和适应性．