面向网络空间防御的对抗机器学习研究综述

机器学习以强大的自适应性和自学习能力成为网络空间防御的研究热点和重要方向. 然而机器学习模型在网络空间环境下存在受到对抗攻击的潜在风险, 可能成为防御体系中最为薄弱的环节, 从而危害整个系统的安全. 为此科学分析安全问题场景, 从运行机理上探索算法可行性和安全性, 对运用机器学习模型构建网络空间防御系统大有裨益. 全面综述对抗机器学习这一跨学科研究领域在网络空间防御中取得的成果及以后的发展方向. 首先, 介绍了网络空间防御和对抗机器学习等背景知识; 其次, 针对机器学习在网络空间防御中可能遭受的攻击, 引入机器学习敌手模型概念, 目的是科学评估其在特定威胁场景下的安全属性; 然后, 针对网络空间防御的机器学习算法, 分别论述了在测试阶段发动规避攻击、在训练阶段发动投毒攻击、在机器学习全阶段发动隐私窃取的方法, 进而研究如何在网络空间对抗环境下, 强化机器学习模型的防御方法; 最后, 展望了网络空间防御中对抗机器学习研究的未来方向和有关挑战.     

N-gram统计模型在机器翻译系统中的应用

文章提出了N-gram模型在机器翻译系统中的几个应用。模型是在语料库的基础上统计连续几个词的出现概率,以此来筛选翻译过程中的侯选元素,并可以对译文的语序进行纠正。由于此种方法是建立在语料库的基础之上的,从而具有真实可靠和实时等特点。实验表明,这种方法具有良好的性能,且与被处理的语言无关。     

机器学习的主要策略综述

当前人工智能研究的主要障碍和发展方向之一就是机器学习。机器学习与计算机科学、心理学、认知科学等各种学科都有着密切的联系,牵涉的面比较广,许多理论及技术上的问题尚处于研究之中。对机器学习的一些主要策略的基本思想进行了较全面的介绍,同时介绍了一些最新的进展和研究热点。     

结合自集成和对抗学习的域自适应城市场景语义分割

针对城市场景标签获取的高额成本问题,文中提出结合自集成和对抗学习的域自适应城市场景语义分割方法.对于源域和目标域的较大域间差异问题,采用风格转换的方法将源域数据集合成具有目标域风格的新数据集,作为新的源域数据集,从而有效减少源域与目标域的域间差异.对于目标域的域内差异问题,引入自集成方法,构造教师网络,利用教师网络在目...     

A literature survey on robust and efficient eye localization in real-life scenarios

Eye localization has gained a wide range of applications in face recognition, gaze estimation, pose estimation, expression analysis, etc. However, due to the high degree of appearance variability of eyes in size, shape, color, texture and various ambient environment changes, this task is challenging. During the last three decades, numerous techniques have been developed to meet these challenges. The goal of this paper is to categorize and evaluate these algorithms in a comprehensive way. We focus on the overall difficulties and challenges in real-life scenarios, and present a detailed review of prominent algorithms from the perspective of learning generalizable, flexible and efficient statistical eye models from a small number of training images. In addition, we organize the discussion of the global aspects of eye localization in uncontrolled environments, towards the development of a robust eye localization system. This paper concludes with several promising directions for future research.     

介词短语识别中规则与统计方法融合的探讨

在介词短语的识别中,统计学习方法和人工规则方法是人们常采用的两种最主要的方法。在浅层句法分析层次上,基于几种典型模式探讨分析构建介词短语识别模型时,规则方法和统计学习方法的有效结合。指出介词短语特征的提取实质是基于语料的语用规则的一种抽象。提出统计学习方法和人工规则方法的有机结合是未来的发展方向。     

一种多重支持向量机在线学习算法的研究

增量式支持向量机学习算法是一种重要的在线学习方法。传统的单增量支持向量机学习算法使用一个数据样本更新支持向量机模型。在增加或删除的数据样本点较多时,这种模型更新模式耗时巨大,具体原因是每个被插入或删除的样本都要进行一次模型参数更新的判断。该文提出一种基于参数规划的多重增量式的支持向量机优化训练算法,使用该训练算法,多重的支持向量机的训练时间大为减少。在合成数据集及真实测试数据集上的实验结果显示,该文提出的方法可以大大降低多重支持向量机训练算法的计算复杂度并提高分类器的精度。     

Generating American Sign Language animation: overcoming misconceptions and technical challenges

Misconceptions about the English literacy rates of deaf Americans, the linguistic structure of American Sign Language (ASL), and the suitability of traditional machine translation (MT) technology to ASL have slowed the development of English-to-ASL MT systems for use in accessibility applications. This article traces the progress of a new English-to-ASL MT project targeted to translating texts important for literacy and user-interface applications. These texts include ASL phenomena called “classifier predicates.” Challenges in producing classifier predicates, novel solutions to these challenges, and applications of this technology to the design of user-interfaces accessible to deaf users will be discussed.     

正则化超限学习机的多分块松弛交替方向乘子法

针对超限学习机在大数据环境下计算负担过重的问题,文中提出正则化超限学习机的多分块松弛交替方向乘子法及N-等分和N/2-等分情形的标量化实现.模型分块使算法具有高度的并行结构,与松弛技术结合提高算法的收敛速度.通过分析,建立算法收敛的充要条件,给出最优收敛率及最优参数.在基准数据集上仿真计算收敛率随分块数的变化关系,对比不同算法的收敛速率和GPU加速比.实验表明,文中算法具有较低的计算复杂度和较高的并行性.     

面向不平衡数据流的自适应加权在线超限学习机算法

一般的在线学习算法对不平衡数据流的分类识别会遇到较大困难,特别是当数据流发生概念漂移时,对其进行分类会变得更困难.文中提出面向不平衡数据流的自适应加权在线超限学习机算法,自动调整实时到达的训练样本的惩罚参数,达到在线学习不平衡数据流的目的.文中算法可以适用于不同偏斜程度的静态数据流的在线学习和发生概念漂移时数据流的在线学习.理论分析和在多个真实数据流上的实验表明文中算法的正确性和有效性.     

基于支持向量机的学习资源推送方案

文章将支持向量机用于学习资源的个性化推送。通过引入知识点的概念与形式化描述,建立起测试试题与学习资源的联系,并形成特征向量。采用线性核函数、多项式核函数和径向基核函数分别构建不同的支持向量机,进行了实验与数据分析。结果表明,支持向量机技术能较好的支持学习资源的个性化推送,具有响应速度快,正确率高,训练样本小等优点。     

深度学习在自然语言处理领域的研究进展

通过定量与定性相结合的方式全面分析了深度学习在自然语言处理领域的研究情况。采用CiteSpace和VOSviewer对深度学习在自然语言处理领域的研究国家、机构、期刊分布、关键词共现、共被引网络聚类及时间轴视图等进行知识图谱绘制,理清研究脉络。通过深入挖掘领域内的重要文献,总结深度学习在自然语言处理领域的研究趋势、存在的主要问题或发展瓶颈,并给出相应的解决办法与思路。对于如何跟踪深度学习在自然语言处理领域的研究成果给出建议,为该领域的后续研究与发展提供参考。     

基于矫正理解的中文文本对抗样本生成方法

自然语言处理技术在文本分类、文本纠错等任务中表现出强大性能,但容易受到对抗样本的影响,导致深度学习模型的分类准确性下降。防御对抗性攻击是对模型进行对抗性训练,然而对抗性训练需要大量高质量的对抗样本数据。针对目前中文对抗样本相对缺乏的现状,提出一种可探测黑盒的对抗样本生成方法 WordIllusion。在数据处理与计算模块中,数据在删除标点符号后输入文本分类模型得到分类置信度,再将分类置信度输入CKSFM计算函数,通过计算比较cksf值选出句子中的关键词。在关键词替换模块中,利用字形嵌入空间和同音字库中的相似词语替换关键词并构建对抗样本候选序列,再将序列重新输入数据处理与计算模块计算cksf值,最终选择cksf值最高的数据作为最终生成的对抗样本。实验结果表明,WordIllusion方法生成的对抗样本在多数深度学习模型上的攻击成功率高于基线方法,在新闻分类场景的DPCNN模型上相比于CWordAttack方法最多高出41.73个百分点,且生成的对抗样本与原始文本相似度很高,具有较强的欺骗性与泛化性。     

A survey on GANs for computer vision: Recent research,analysis and taxonomy

In the last few years, there have been several revolutions in the field of deep learning, mainly headlined by the large impact of Generative Adversarial Networks (GANs). GANs not only provide an unique architecture when defining their models, but also generate incredible results which have had a direct impact on society. Due to the significant improvements and new areas of research that GANs have brought, the community is constantly coming up with new researches that make it almost impossible to keep up with the times. Our survey aims to provide a general overview of GANs, showing the latest architectures, optimizations of the loss functions, validation metrics and application areas of the most widely recognized variants. The efficiency of the different variants of the model architecture will be evaluated, as well as showing the best application area; as a vital part of the process, the different metrics for evaluating the performance of GANs and the frequently used loss functions will be analyzed. The final objective of this survey is to provide a summary of the evolution and performance of the GANs which are having better results to guide future researchers in the field.     

基于极限学习机的XML文档分类

研究基于极限学习机(ELM)的XML文档分类方法。为优化文档的相似性计算,在结构链接向量模型的基础上,提出一种改进的特征向量模型RS-VSM,将有效的结构化信息合并到向量模型中。应用ELM对XML文档进行分类,为提高ELM分类的准确率,提出一种基于投票机制的Voting-ELM算法。实验结果证明,该算法的分类效果较优。     

基于语种关联度课程学习的多语言神经机器翻译

近年来,使用单一模型实现多语言神经机器翻译的方法受到了广泛关注.然而,现有方法多将所有语种语料直接混合作为训练语料,未能利用多种语言之间关联和相似的信息.此外,模型训练涉及语言种类多、数据量大、整体训练难度大、耗时长等问题.针对以上两个问题,文中提出了一种基于语种关联度的课程学习方法来提高多语言神经机器翻译的整体性能和...     

A Semantic Network of English: The Mother of All WordNets

We give a brief outline of the design and contents of the English lexical database WordNet, which serves as a model for similarly conceived wordnets in several European languages. WordNet is a semantic network, in which the meanings of nouns, verbs, adjectives, and adverbs are represented in terms of their links to other (groups of) words via conceptual-semantic and lexical relations. Each part of speech is treated differently reflecting different semantic properties. We briefly discuss polysemy in WordNet, and focus on the case of meaning extensions in the verb lexicon. Finally, we outline the potential uses of WordNet not only for applications in natural language processing, but also for research in stylistic analyses in conjunction with a semantic concordance.     

步态识别的深度学习:综述

由于步态容易受到物体遮挡、衣着、视角和携带物等协变量因素的影响,步态识别方法较难获得较优的识别性能.基于端到端和多层特征提取的思想,深度学习近年在步态识别领域取得一系列进展.本文综述深度学习在步态识别中的研究现状、优势和不足,总结其中的关键技术和潜在的研究方向.     

A Novel Multi-classifier Integrated Model for Chinese Noun Sense Disambiguation

We propose a Multi-classifier Compatible Computational Model （MCCM） for Chinese noun sense disambiguation in this paper. In natural language processing, many problems can be viewed as classification in nature. However, different classifiers cannot be efficiently integrated by a general standard. Quantification rule is introduced for flexible integration of statistical and role-based classifiers. Rules are traced into the corpus for acquiring their quantification information. The other contribution is Projected Performance Evaluation Matrix （P-PEM）. Classification performance is improved because it contains more accurate classification information for every classifier. Many faint classifiers boost a powerful MCCM model. Finally the experiment shows its advantages.     

Enhancing computer graphics through machine learning: a survey

Machine learning has experienced explosive growth in the last few decades, achieving sufficient maturity to provide effective tools for sundry scientific and engineering fields. Machine learning provides a firm theoretical foundation upon which to build techniques that leverage existing data to extract interesting information or to synthesize more data.In this paper we survey the uses of machine learning methods and concepts in recent computer graphics techniques. Many graphics techniques are data-driven; however, few graphics papers explicitly leverage the machine learning literature to underpin, validate, and develop their proposed methods. This survey provides novel insights by casting many existing computer graphics techniques into a common learning framework. This not only illuminates how these techniques are related, but also reveals possible ways in which they may be improved. We also use our analysis to propose several directions for future work.