内容提要 大规模多媒体信息管理与检索面临着两大类艰巨的技术挑战。首先,这一工程问题的研究在本质上是多领域、跨学科的,涉及信号处理、计算机视觉、数据库、机器学习、神经科学和认知心理学;其次,一个有效的解决方案必须能解决高维数据和网络规模数据的可扩展性问题。《大规模多媒体信息管理与检索基础(英):模拟人类感知数学方法》部分(~8章)着重介绍如何采用多领域、跨学科算法来解决特征提取及选择、知识表示、语义分析、距离函数的制定等问题;第二部分(第9~12章)对解决高维数据和网络规模数据的扩展性问题提出了有效的处理方法。此外,《大规模多媒体信息管理与检索基础(英):模拟人类感知数学方法》的附录还给出了作者开发的开源软件的下载地址。 《大规模多媒体信息管理与检索基础(英):模拟人类感知数学方法》是作者在美国加州大学从事多年的教学科研及在google公司工作多年的基础上编写的。《大规模多媒体信息管理与检索基础(英):模拟人类感知数学方法》适合多媒体、计算机视觉、机器学习、大规模数据处理等领域的研发人员阅读,也可作为高等院校计算机专业本科生及研究生的教材或教学参考书。 目录 1 introduction - key subroutines of multimedia datamanagement1.1 overview1.2 feature extractio1.3 similarity1.4 learning1.5 multimodal fusio1.6 indeng1.7 scalability1.8 concluding remarksreferences2 perceptual feature extractio2.1 introductio2.2 dmd algorithm2.2.1 model-based pipeline2.2.2 data-driven pipeline2.3 experiments2.3.1 dataset and setup2.3.2 model-based vs. data-drive2.3.3 dmd vs. individual models2.3.4 regularization tuning2.3.5 tough categories2.4 related reading2.5 concluding remarksreferences3 query concept learning3.1 introductio3.2 support vector machines and version space3.3 active learning and batch sampling strategies3.3.1 theoretical foundatio3.3.2 sampling strategies3.4 concept-dependent learning3.4.1 concept complety3.4.2 limitations of active learning3.4.3 concept-dependent active learning algorithms3.5 experiments and discussio3.5.1 testbed and setup3.5.2 active vs. passive learning3.5.3 against traditional relevance feedback schemes3.5.4 sampling method evaluatio3.5.5 concept-dependent learning3.5.6 concept diversity evaluatio3.5.7 evaluation summary3.6 related reading3.6.1 machine learning3.6.2 relevance feedback3.7 relation to other chapters3.8 concluding remarksreferences4 similarity4.1 introductio4.2 mining image feature set4.2.1 image testbed setup4.2.2 feature extractio4.2.3 feature selectio4.3 discovering the dynamic partial distance functio4.3.1 minkowski metric and its limitations4.3.2 dynamic partial distance functio4.3.3 psychological interpretation of dynamic partial distancefunctio4.4 empirical study4.4.1 image retrieval4.4.2 video shot-transition detectio4.4.3 near duplicated articles4.4.4 weighted dpf vs. weighted euclidea4.4.5 observations4.5 related reading4.6 concluding remarksreferences5 formulating distance functions5.1 introductio5.2 dfa algorithm5.2.1 transformation model5.2.2 distance metric learning5.3 experimental evaluatio5.3.1 evaluation on contextual informatio5.3.2 evaluation on effectiveness5.3.3 observations5.4 related reading5.4.1 metric learning5.4.2 kernel learning5.5 concluding remarksreferences6 multimodal fusio6.1 introductio6.2 related reading6.2.1 modality identificatio6.2.2 modality fusio6.3 independent modality analysis6.3.1 pca6.3.2 ica6.3.3 img6.4 super-kernel fusio6.5 experiments6.5.1 evaluation of modality analysis6.5.2 evaluation of multimodal kernel fusio6.5.3 observations6.6 concluding remarksreferences7 fusing content and context with causality7.1 introductio7.2 related reading7.2.1 photo annotatio7.2.2 probabilistic graphical models7.3 multimodal metadata7.3.1 contextual informatio7.3.2 perceptual content7.3.3 semantic ontology7.4 influence diagrams7.4.tructure learning7.4.2 causal strength7.4.3 case study7.4.4 dealing with missing attributes7.5 experiments7.5.1 experiment on learning structure7.5.2 experiment on causal strength inference7.5.3 experiment on semantic fusio7.5.4 experiment on missing features7.6 concluding remarksreferences8 combinational collaborative filtering, consideringpersonalizafio8.1 introductio8.2 related reading8.3 ccf: combinational collaborative filtering8.3.1 c-u and c-d baseline models8.3.2 ccf model8.3.3 gibbs & em hybrid training8.3.4 parallelizatio8.3.5 inference8.4 experiments8.4.1 gibbs em vs. em8.4.2 the orkut dataset8.4.3 runtime speedup8.5 concluding remarksreferences9 imbalanced data learning9.1 introductio9.2 related reading9.3 kernel boundary alignment9.3.1 conformally transforming kernel k9.3.2 modifying kernel matrix k9.4 experimental results9.4.1 vector-space evaluatio9.4.2 non-vector-space evaluatio9.5 concluding remarksreferences10 psvm: parallelizing support vector machines on distributedcomputers10.1 introductio10.2 interior point method with incomplete choleskyfactorizatio10.3 psvm algorithm10.3.1 parallel icf10.3.2 parallel ipm10.3.3 computing parameter b and writing back10.4 experiments10.4.1 class-prediction accuracy10.4.2 scalability10.4.3 overheads10.5 concluding remarksreferences11 appromate high-dimensional indeng with kernel11.1 introductio11.2 related reading11.3 algorithm spheredex11.3.1 create - building the index11.3.2 search - querying the index11.3.3 update - insertion and deletio11.4 experiments11.4.etup11.4.2 performance with disk ios11.4.3 choice of parameter g11.4.4 impact of insertions11.4.5 sequential vs. random11.4.6 percentage of data processed11.4.7 summary11.5 concluding remarks11.5.1 range queries11.5.2 farthest neior queriesreferences12 speeding up latent dirichlet allocation with parallelization andpipeline strategies12.1 introductio12.2 related reading12.3 ad-lda: appromate distributed lda12.3.1 parallel gibbs sampling and allreduce12.3.2 mpi implementation of ad-lda12.4 plda 12.4.1 reduce bottleneck of ad-lda12.4.2 framework of plda 12.4.3 algorithm for pw processors12.4.4 algorithm for pd processors12.4.5 straggler handling12.4.6 parameters and complety12.5 experimental results12.5.1 datasets and experiment environment12.5.2 perplety12.5.3 speedups and scalability12.6 large-scale applications12.6.1 mining social-network user latent behavior12.6.2 question labeling (ql)12.7 concluding remarksreferences 作者介绍 Feature extraction is fundamental to all multimedia computing tasks. Features can be classified into two categories, content and context. Content refers directly to raw imagery, video, and mucic data such as pixels, motions, and tones, respectively, and their representations. Context refers to metadata collected or associated withcontent when a piece of data is acquired or published. For instance, EXIF cameraparameters and GPS location are contextual information that some digital camerascan collect. Other widely used contextual information includes surrounding texts ofan image/photo on a Web page, and social interactions on a piece of multimediadata instance. Context and content ought to be fused synergistically when analyzingmultimedia data. …… 序言
以下为对购买帮助不大的评价