深度学习及其在目标和行为识别中的新进展

深度学习是机器学习中的一个新的研究领域。通过深度学习的方法构建深度网络来抽取特征是目前目标和行为识别中得到关注的研究方向。为引起更多计算机视觉领域研究者对深度学习进行探索和讨论，并推动目标和行为识别的研究，本文对深度学习及其在目标和行为识别中的新进展给予了概述。本文先介绍深度学习领域研究的基本状况、主要概念和原理；然后介绍近期利用深度学习在目标和行为识别应用中的一些新进展；最后阐述了深度学习与神经网络之间的关系，深度学习的优缺点，以及目前深度学习理论需要解决的主要问题。这对拟将深度学习应用于目标和行为识别的研究人员应有所帮助。  

大数据系统和分析技术综述

首先根据处理形式的不同,介绍了不同形式数据的特征和各自的典型应用场景以及相应的代表性处理系统,总结了大数据处理系统的三大发展趋势;随后,对系统支撑下的大数据分析技术和应用(包括深度学习、知识计算、社会计算与可视化等)进行了简要综述,总结了各种技术在大数据分析理解过程中的关键作用;最后梳理了大数据处理和分析面临的数据复杂性、计算复杂性和系统复杂性挑战,并逐一提出了可能的应对之策.  

深度学习研究进展

深度学习(Deep Learning)是一个近几年备受关注的研究领域,在机器学习中起着重要的作用.如果说浅层学习是机器学习的一次浪潮,那么深度学习作为机器学习的一个新领域,将掀起机器学习的又一次浪潮.深度学习通过建立、模拟人脑的分层结构来实现对外部输入的数据进行从低级到高级的特征提取,从而能够解释外部数据.首先介绍了深度学习的由来,分析了浅层学习存在的弊端;其次列举了深度学习的经典方法,主要以监督学习和无监督学习来展开介绍;然后对深度学习的最新研究进展及其应用进行了综述;最后总结了深度学习发展所面临的问题.  

深度学习研究综述

深度学习是一类新兴的多层神经网络学习算法,因其缓解了传统训练算法的局部最小性,引起机器学习领域的广泛关注。首先论述了深度学习兴起渊源,分析了算法的优越性,并介绍了主流学习算法及应用现状,最后总结了当前存在的问题及发展方向。  

不同池化模型的卷积神经网络学习性能研究

目的 基于卷积神经网络的深度学习算法在图像处理领域正引起广泛关注。为了进一步提高卷积神经网络特征提取的准确度，加快参数收敛速度，优化网络学习性能，通过对比不同的池化模型对学习性能的影响提出一种动态自适应的改进池化算法。方法 构建卷积神经网络模型，使用不同的池化模型对网络进行训练，并检验在不同迭代次数下的学习结果。在现有算法准确率不高和收敛速度较慢的情况下，通过使用不同的池化模型对网络进行训练，从而构建一种新的动态自适应池化模型，并研究在不同迭代次数下其对识别准确率和收敛速度的影响。结果 通过对比实验发现，使用动态自适应池化算法的卷积神经网络学习性能最优，在手写数字集上的收敛速度最高可以提升18.55%，而模型对图像的误识率最多可以降低20%。结论 动态自适应池化算法不但使卷积神经网络对特征的提取更加精确，而且很大程度地提高了收敛速度和模型准确率，从而达到优化网络学习性能的目的。这种模型可以进一步拓展到其他与卷积神经网络相关的深度学习算法。  

Engaging online learners: The impact of Web-based learning technology on college student engagement

Widespread use of the Web and other Internet technologies in postsecondary education has exploded in the last 15 years. Using a set of items developed by the National Survey of Student Engagement (NSSE), the researchers utilized the hierarchical linear model (HLM) and multiple regressions to investigate the impact of Web-based learning technology on student engagement and self-reported learning outcomes in face-to-face and online learning environments. The results show a general positive relationship between the use the learning technology and student engagement and learning outcomes. We also discuss the possible impact on minority and part-time students as they are more likely to enroll in online courses.  

Examining online teaching,cognitive, and social presence for adult students

Drawing on the Community of Inquiry model (Garrison, Anderson, & Archer, 2000), this mixed-method case study examined the nature and interactions of teaching, cognitive, and social presence created by online instructors and adult students in diverse course contexts. The study results indicated online instructional design and teaching elements that are crucial prerequisites for a successful online higher educational experience for adult students. The study also informed e-learning designers on the relations between online teaching, cognitive, and social presence.  

Exploring the characteristics of an optimal design for inquiry-based geography education with Geographic Information Systems

Geographic Information Systems (GIS) is a kind of computer software that allows people to work with digital maps in a fast and flexible way. In the past decade, more and more geography teachers have become interested in the possibilities of using GIS in secondary education. However, teaching with GIS is complex, and little is known about how to do so in an optimal way. Therefore, an Educational Design Research study (EDR) was conducted with the aim to explore the characteristics of an optimal design for GIS-supported geographic inquiry projects. In this EDR study, a project was developed via progressive cycles of designing, testing, and evaluating, together with teachers from different schools. This paper summarizes the outcomes of the EDR study, and presents some design principles for GIS-supported inquiry-based geography education. Teachers could use these design principles to design and conduct GIS-supported geographic inquiry projects, and in such a way raise their geography lessons to a higher level.This paper also shows that although GIS provides many opportunities for enhancing inquiry-based geography projects, it also holds many conditions for its use to be optimal. GIS-supported inquiry-based geography education requires more than providing appropriate software, tasks, and coaching to ensure that students do not get stuck. In order to effectively raise students’ geographic thinking to a higher level, the project should offer a considerable amount of guidance: it should include several preparatory and evaluative tasks based on a good domain-specific theory for use in educational settings. In addition, teachers should coach students in structuring, correcting, and expanding their geographic thinking via dialogical teaching.  

深度学习的昨天、今天和明天

机器学习是人工智能领域的一个重要学科.自从20世纪80年代以来,机器学习在算法、理论和应用等方面都获得巨大成功.2006年以来,机器学习领域中一个叫“深度学习”的课题开始受到学术界广泛关注,到今天已经成为互联网大数据和人工智能的一个热潮.深度学习通过建立类似于人脑的分层模型结构,对输入数据逐级提取从底层到高层的特征,从而能很好地建立从底层信号到高层语义的映射关系.近年来,谷歌、微软、IBM、百度等拥有大数据的高科技公司相继投入大量资源进行深度学习技术研发,在语音、图像、自然语言、在线广告等领域取得显著进展.从对实际应用的贡献来说,深度学习可能是机器学习领域最近这十年来最成功的研究方向.将对深度学习发展的过去和现在做一个全景式的介绍,并讨论深度学习所面临的挑战,以及将来的可能方向.  

基于受限波兹曼机的推荐算法研究

本文针对推荐系统中传统推荐算法在处理较稀疏数据效果表现不佳的问题，将一种最新的机器学习方法带入到推荐算法中，利用多层波兹曼机组成的深度结构模型与传统最近邻推荐方法相结合，形成一种新的推荐模型算法。本文还利用有限步吉布斯采样的最小化散度差（Constrastive Divergence）方法解决了该模型的无监督训练问题，并且通过预训练和反馈微调使得模型的训练得以实现；最后利用深度结构抽取的抽象特征结合最近邻方法进行预测推荐。另外，本文利用传统基于相似度最近邻方法，矩阵分解方法和新模型算法在相关数据集上进行多组实验，实验结果表明该算法不仅在稀疏的数据上也表现出更好的效果，并且拥有着更快的收敛速度。