Part 2: Multilayer perceptron and natural gradient learning

Since the perceptron was developed for learning to classify input patterns, there have been plenty of studies on simple perceptrons and multilayer perceptrons. Despite wide and active studies in theory and applications, multilayer perceptrons still have many unsettled problems such as slow learning speed and overfitting. To find a thorough solution to these problems, it is necessary to consolidate previous studies, and find new directions for uplifting the practical power of multilayer perceptrons. As a first step toward the new stage of studies on multilayer perceptrons, we give short reviews on two interesting and important approaches; one is stochastic approach and the other is geometric approach. We also explain an efficient learning algorithm developed from the statistical and geometrical studies, which is now well known as the natural gradient learning method. Hyeyoung Park, Ph.D.: She is Assistant Professor of Computer Sciences at School of Electrical Engineering and Computer Science of Kyungpook National University in Korea. She received her B.S., M.A. and Ph.D. from Yonsei University of Korea in 1994, 1996, and 2000. She also worked as a research scinetist at Brain Science Institute in RIKEN from 2000 to 2004. Her research insterest is in learning thoeries and pattern recognition as well as statistical data analysis.     

Adaptive method of realizing natural gradient learning for multilayer perceptrons

The natural gradient learning method is known to have ideal performances for on-line training of multilayer perceptrons. It avoids plateaus, which give rise to slow convergence of the backpropagation method. It is Fisher efficient, whereas the conventional method is not. However, for implementing the method, it is necessary to calculate the Fisher information matrix and its inverse, which is practically very difficult. This article proposes an adaptive method of directly obtaining the inverse of the Fisher information matrix. It generalizes the adaptive Gauss-Newton algorithms and provides a solid theoretical justification of them. Simulations show that the proposed adaptive method works very well for realizing natural gradient learning.     

Part 4: Reinforcement learning: Machine learning and natural learning

The theory of reinforcement learning (RL) was originally motivated by animal learning of sequential behavior, but has been developed and extended in the field of machine learning as an approach to Markov decision processes. Recently, a number of neuroscience studies have suggested a relationship between reward-related activities in the brain and functions necessary for RL. Regarding the history of RL, we introduce in this article the theory of RL and present two engineering applications. Then we discuss possible implementations in the brain.     

Consciousness-driven reinforcement learning: An online learning control framework

As a powerful tool for solving nonlinear complex system control problems, the model-free reinforcement learning hardly guarantees system stability in the early stage of learning, especially with high complicity learning components applied. In this paper, a reinforcement learning framework imitating many cognitive mechanisms of brain such as attention, competition, and integration is proposed to realize sample-efficient self-stabilized online learning control. Inspired by the generation of consciousness in human brain, multiple actors that work either competitively for best interaction results or cooperatively for more accurate modeling and predictions were applied. A deep reinforcement learning implementation for challenging control tasks and a real-time control implementation of the proposed framework are respectively given to demonstrate the high sample efficiency and the capability of maintaining system stability in the online learning process without requiring an initial admissible control.     

Interactive online learning

This paper describes a system for visual object recognition based on mobile augmented reality gear. The user can train the system to the recognition of objects online using advanced methods of interaction with mobile systems: Hand gestures and speech input control “virtual menus,” which are displayed as overlays within the camera image. Here we focus on the underlying neural recognition system, which implements the key requirement of an online trainable system—fast adaptation to novel object data. The neural three-stage architecture can be adapted in two modes: In a fast training mode (FT), only the last stage is adapted, whereas complete training (CT) rebuilds the system from scratch. Using FT, online acquired views can be added at once to the classifier, the system being operational after a delay of less than a second, though still with reduced classification performance. In parallel, a new classifier is trained (CT) and loaded to the system when ready. The text was submitted by the authors in English. Gunther Heidemann was born in 1966. He studied physics at the Universities of Karlsruhe and Münster and received his PhD (Eng.) from Bielefeld University in 1998. He is currently working within the collaborative research project “Hybrid Knowledge Representation” of the SFB 360 at Bielefeld University. His fields of research are mainly computer vision, robotics, neural networks, data mining, bonification, and hybrid systems. Holger Bekel was born in 1970. He received his BS degree from the University of Bielefeld, Germany, in 1997. In 2002 he received a diploma in Computer Science from the University of Bielefeld. He is currently pursuing a PhD program in Computer Science at the University of Bielefeld, working within the Neuroinformatics Group (AG Neuroinformatik) in the project VAMPIRE (Visual Active Memory Processes and Interactive Retrieval). His fields of research are active vision and data mining. Ingo Bax was born in 1976. He received a diploma in Computer Science from the University of Bielefeld in 2002. He is currently pursuing a PhD program in Computer Science at the Neuroinformatics Group of the University of Bielefeld, working within the VAMPIRE project. His fields of interest are cognitive computer vision and pattern recognition. Helge J. Ritter was born 1958. He studied physics and mathematics at the Universities of Bayreuth, Heidelberg and Munich. After a PhD in physics at Technical University of Munich in 1988, he visited the Laboratory of Computer Science at Helsinki University of Technology and the Beckman Institute for Advanced Science and Technology at the University of Illinois at Urbana-Champaign. Since 1990 he has headed the Neuroinformatics Group at the Faculty of Technology, Bielefeld University. His main interests are principles of neural computation and their application to building intelligent systems. In 1999, she was awarded the SEL Alcatel Research Prize, and in 2001, the Leibniz Prize of the German Research Foundation DFG.     

Improving generalization performance of natural gradient learning using optimized regularization by NIC

Natural gradient learning is known to be efficient in escaping plateau, which is a main cause of the slow learning speed of neural networks. The adaptive natural gradient learning method for practical implementation also has been developed, and its advantage in real-world problems has been confirmed. In this letter, we deal with the generalization performances of the natural gradient method. Since natural gradient learning makes parameters fit to training data quickly,the overfitting phenomenon may easily occur, which results in poor generalization performance. To solve the problem, we introduce the regularization term in natural gradient learning and propose an efficient optimizing method for the scale of regularization by using a generalized Akaike information criterion (network information criterion). We discuss the properties of the optimized regularization strength by NIC through theoretical analysis as well as computer simulations. We confirm the computational efficiency and generalization performance of the proposed method in real-world applications through computational experiments on benchmark problems.     

Assessing online collaborative learning: process and product

The assessment of online collaborative study presents new opportunities and challenges, both in terms of separating the process and product of collaboration, and in the support of skills development. The purpose of this paper is to explore the role of assessment with respect to the processes and products of online collaborative study. It describes a qualitative case study of staff and students perspectives on two UK Open University courses which have used a variety of models of online collaborative assessment. The findings underline the importance of assessment in ensuring online participation, and in supporting the practice and development of online collaborative learning. They have led to a number of recommendations for the assessment of online collaborative learning.     

Evaluating online solutions for experiential support of distance learning by teachers in China

This paper describes and assesses the development of an online solution for the experiential support of distance learning by teachers. Three hundred and forty-eight randomly selected K-12 teachers participated in this pilot study using the online learning environment designed in this research. Teachers' products, surveys, and interviews were collected and analysed. Results showed that the teacher-learners could learn as well as in face-to-face learning in an earlier implementation of the course. The learning support system as designed fits teacher-learners' needs. They benefited from learning communities formed online as well as face-to-face. More support for online discussion and example cases are needed to support experiential learning. Suggestions are made to improve the design of the learning support system and the pedagogy for experiential teacher learning.     

Implementing projection pursuit learning

This paper examines the implementation of projection pursuit regression (PPR) in the context of machine learning and neural networks. We propose a parametric PPR with direct training which achieves improved training speed and accuracy when compared with nonparametric PPR. Analysis and simulations are done for heuristics to choose good initial projection directions. A comparison of a projection pursuit learning network with a single hidden-layer sigmoidal neural network shows why grouping hidden units in a projection pursuit learning network is useful. Learning robot arm inverse dynamics is used as an example problem.     

System-in-package testing: problems and solutions

System-in-package integrates multiple dies in a common package. Therefore, testing SiP technology is different from system-on-chip, which integrates multiple vendor parts. This article provides test strategies for known good die and known good substrate in the SiP. Case studies prove feasibility using the IEEE 1500 test structure.     

Group online adaptive learning

Sharing information among multiple learning agents can accelerate learning. It could be particularly useful if learners operate in continuously changing environments, because a learner could benefit from previous experience of another learner to adapt to their new environment. Such group-adaptive learning has numerous applications, from predicting financial time-series, through content recommendation systems, to visual understanding for adaptive autonomous agents. Here we address the problem in the context of online adaptive learning. We formally define the learning settings of Group Online Adaptive Learning and derive an algorithm named Shared Online Adaptive Learning (SOAL) to address it. SOAL avoids explicitly modeling changes or their dynamics, and instead shares information continuously. The key idea is that learners share a common small pool of experts, which they can use in a weighted adaptive way. We define group adaptive regret and prove that SOAL maintains known bounds on the adaptive regret obtained for single adaptive learners. Furthermore, it quickly adapts when learning tasks are related to each other. We demonstrate the benefits of the approach for two domains: vision and text. First, in the visual domain, we study a visual navigation task where a robot learns to navigate based on outdoor video scenes. We show how navigation can improve when knowledge from other robots in related scenes is available. Second, in the text domain, we create a new dataset for the task of assigning submitted papers to relevant editors. This is, inherently, an adaptive learning task due to the dynamic nature of research fields evolving in time. We show how learning to assign editors improves when knowledge from other editors is available. Together, these results demonstrate the benefits for sharing information across learners in concurrently changing environments.

     

Research in online learning environments: Priorities and methodologies

Due to increasing demand for education and training in the information age, online learning and teaching is becoming a necessity in our future. However, lack of research goals to understand impact of online learning environments on students is a problem in research on online learning environments. We identified four main research goals to pursue in online learning environments based on their impact on learner achievement, engagement, and retention (opposite of attrition). Those goals are (a) enhancing learner engagement & collaboration, (b) promoting effective facilitation, (c) developing assessment techniques, and (d) designing faculty development programs. Current promising work in those areas is presented. Four methods that are common in the instructional technology literature are recommended to pursue those goals. Formative research and developmental research are relevant for all four. Although appropriate for any of the goals, experimental research is a better fit for goals b and c, and activity theory is useful for goals a and b.     

A theory of online learning as online participation

In this paper, an initial theory of online learning as online participation is suggested. It is argued that online learner participation (1) is a complex process of taking part and maintaining relations with others, (2) is supported by physical and psychological tools, (3) is not synonymous with talking or writing, and (4) is supported by all kinds of engaging activities. Participation and learning are argued to be inseparable and jointly constituting. The implication of the theory is straightforward: If we want to enhance online learning, we need to enhance online learner participation.     

多层次网络学习模型与学习控制研究

在网络化的学习环境中,由于存在着学习需求多样化和缺乏针对性学习内容情况,因此如何提供一套完整的学习模型和学习控制机制对于提高学生学习效率显得尤为重要。基于知识空间理论和高级Petri网技术,建立了基于同一学习内容的,适应不同学习需求和不同学习对象的多层次的学习模型和学习控制机制,通过该模型可以很大程度上满足以上要求,避免了“知识迷路”和“学习迷航”,提供更加个性化的学习指导。     

一种基于簇的极限学习机的在线学习算法

针对传统的批量学习算法学习速度慢、对空间需求量高的缺点,提出了一种基于簇的极限学习机的在线学习算法。该算法将分簇的理念融入到极限学习机中,并结合极限学习机,提出了一种基于样本类别和样本输出的分簇标准;同时提出了一种加权的Moore-Penrose算法求隐层节点与输出节点的连接权重。实验结果表明,该算法具有学习能力好、拟合度高、泛化性能好等优点。     

A framework for online learning: the Virtual-U

Much of the online post-secondary education available in North America and Europe has been created piecemeal. This situation arose because educators began adopting computer networking in the mid-1970s, soon after the invention of packet-switched networks (1969) and e-mail and computer conferencing (1971) for exchange of scientific information. In late 1993, the author set out to help design a system using the Internet that would encourage the adoption of a collaborative learning approach. She and her colleagues also wanted to develop embedded tools to meet the needs of both instructors and students. The goal of their system, now known as the Virtual-U (http://www.vu.vlei.com), was to provide a flexible framework to support advanced pedagogies based on active learning, collaboration, multiple perspectives, and knowledge building. With two years of field trials serving more than 8000 students and hosting 300 courses from 14 institutions, the Virtual-U provides a flexible yet well-organized framework for online, collaborative education. It brings together a multidisciplinary research team of educators, HCI specialists, engineers, computer scientists, and database and instructional designers, fulfilling the promise of integrated online learning     

Exploring dimensions to online learning

The study presented in this paper sought to explore several dimensions to online learning. Identifying the dimensions to online learning entails important basic issues which are of great relevance to educators today. The primary question is “what are the factors that contribute to the success/failure of online learning?” In order to answer this question, we need to identify the important variables that: (1) measure the learning outcome and (2) help us understand the learning experience of students using specific learning tools. In this study, the dimensions we explored are student’s attitude, affect, motivation and perception of an Online Learning Tool usage. A survey methodology was used such that validated items from previous relevant research work were adopted. 105 students completed the questionnaire. An exploratory factor analysis (EFA) was implemented on the data captured. Results of the EFA identified the items that are relevant to the present context and the items that can be used to measure the dimension to online learning. Affect and perception were found to have strong measurement capabilities with the adopted items while motivation was measured the weakest.     

Evaluating peer learning and assessment in online collaborative learning environments

Transformation of learning and teaching in higher education now offers greater educational equality through enhanced access and collaboration within the framework of lifelong learning in the digital age. This study aims to evaluate online peer learning and assessment in the collaborative learning process in higher education practices. The study also investigates the impact of online peer learning on the development of skills within collaborative learning through the use of volunteered responses from learners concerning their experiences with and perceptions of online learning. Therefore, a quantitative approach is applied through the administration of a survey with 32 items that is distributed to 715 participants. According to the objective of the study, a set of inferential statistical analyses are performed. The theoretical framework of this study is the CHAT (cultural historical activity theory) which reconstructs the knowledge of learners through the application of the Adobe Connect program to demonstrate how learners can be collaborative and social with their peers in an online context. The results revealed that the collaborative online peer learning process in higher education encourages critical reflection and self-assessment. The study contributes to the understanding of the value of learner satisfaction in online collaborative learning environments through the experiences of learners.     

The ebb and flow of online learning

Past research has suggested that Csikszentmihalyi's flow theory describes a state that should be supportive of a student's learning. This paper reports on research that uses the constructs of flow to explore learning in an online environment. An experiment was carried out in which students worked through a learning sequence in the physics domain that had varying degrees of interactivity. Their interactions and flow states were monitored throughout the learning task. The experimental data suggest that flow can be more usefully regarded as a process rather than just an overall state. This process is represented by flow-paths that plot each student's progress through challenge-skill space. Some flow patterns are identified that relate to the learning outcomes of the students. While there is some conflict between this process representation and outcome measures for flow, this flow-path portrayal has provided fresh insights into students' interactions in online learning environments.     

Compatible natural gradient policy search

Machine Learning - Trust-region methods have yielded state-of-the-art results in policy search. A common approach is to use KL-divergence to bound the region of trust resulting in a natural...