Edward Y.Chang编著的《大规模多媒体信息管理与检索基础(模拟人类感知数学方法)(精)》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，全书系统全面介绍了大规模多媒体信息管理与检索相关知识，本书适合多媒体、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。

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。Edward Y.Chang编著的《大规模多媒体信息管理与检索基础(模拟人类感知数学方法)(精)》第一部分（第1～8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(模拟人类感知数学方法)(精)》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(模拟人类感知数学方法)(精)》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(模拟人类感知数学方法)(精)》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。

1 introduction - key subroutines of multimedia data management

 1.1 overview

 1.2 feature extraction

 1.3 similarity

 1.4 learning

 1.5 multimodal fusion

 1.6 indexing

 1.7 scalability

 1.8 concluding remarks

 references

2 perceptual feature extraction

 2.1 introduction

 2.2 dmd algorithm

2.2.1 model-based pipeline

2.2.2 data-driven pipeline

 2.3 experiments

2.3.1 dataset and setup

2.3.2 model-based vs. data-driven

2.3.3 dmd vs. individual models

2.3.4 regularization tuning

2.3.5 tough categories

 2.4 related reading

 2.5 concluding remarks

 references

3 query concept learning

 3.1 introduction

 3.2 support vector machines and version space

 3.3 active learning and batch sampling strategies

3.3.1 theoretical foundation

3.3.2 sampling strategies

 3.4 concept-dependent learning

3.4.1 concept complexity

3.4.2 limitations of active learning

3.4.3 concept-dependent active learning algorithms

 3.5 experiments and discussion

3.5.1 testbed and setup

3.5.2 active vs. passive learning

3.5.3 against traditional relevance feedback schemes

3.5.4 sampling method evaluation

3.5.5 concept-dependent learning

3.5.6 concept diversity evaluation

3.5.7 evaluation summary

 3.6 related reading

3.6.1 machine learning

3.6.2 relevance feedback

 3.7 relation to other chapters

 3.8 concluding remarks

 references

4 similarity

 4.1 introduction

 4.2 mining image feature set

4.2.1 image testbed setup

4.2.2 feature extraction

4.2.3 feature selection

 4.3 discovering the dynamic partial distance function

4.3.1 minkowski metric and its limitations

4.3.2 dynamic partial distance function

4.3.3 psychological interpretation of dynamic partial distance function

 4.4 empirical study

4.4.1 image retrieval

4.4.2 video shot-transition detection

4.4.3 near duplicated articles

4.4.4 weighted dpf vs. weighted euclidean

4.4.5 observations

 4.5 related reading

 4.6 concluding remarks

 references

5 formulating distance functions

 5.1 introduction

 5.2 dfa algorithm

5.2.1 transformation model

5.2.2 distance metric learning

 5.3 experimental evaluation

5.3.1 evaluation on contextual information

5.3.2 evaluation on effectiveness

5.3.3 observations

 5.4 related reading

5.4.1 metric learning

5.4.2 kernel learning

 5.5 concluding remarks

 references

6 multimodal fusion

 6.1 introduction

 6.2 related reading

6.2.1 modality identification

6.2.2 modality fusion

 6.3 independent modality analysis

6.3.1 pca

6.3.2 ica

6.3.3 img

 6.4 super-kernel fusion

 6.5 experiments

6.5.1 evaluation of modality analysis

6.5.2 evaluation of multimodal kernel fusion

6.5.3 observations

 6.6 concluding remarks

 references

7 fusing content and context with causality

 7.1 introduction

 7.2 related reading

7.2.1 photo annotation

7.2.2 probabilistic graphical models

 7.3 multimodal metadata

7.3.1 contextual information

7.3.2 perceptual content

7.3.3 semantic ontology

 7.4 influence diagrams

7.4.1 structure learning

7.4.2 causal strength

7.4.3 case study

7.4.4 dealing with missing attributes

 7.5 experiments

7.5.1 experiment on learning structure

7.5.2 experiment on causal strength inference

7.5.3 experiment on semantic fusion

7.5.4 experiment on missing features

 7.6 concluding remarks

 references

8 combinational collaborative filtering, considering personalizafion

 8.1 introduction

 8.2 related reading

 8.3 ccf: combinational collaborative filtering

8.3.1 c-u and c-d baseline models

8.3.2 ccf model

8.3.3 gibbs & em hybrid training

8.3.4 parallelization

8.3.5 inference

 8.4 experiments

8.4.1 gibbs + em vs. em

8.4.2 the orkut dataset

8.4.3 runtime speedup

 8.5 concluding remarks

 references

9 imbalanced data learning

 9.1 introduction

 9.2 related reading

 9.3 kernel boundary alignment

9.3.1 conformally transforming kernel k

9.3.2 modifying kernel matrix k

 9.4 experimental results

9.4.1 vector-space evaluation

9.4.2 non-vector-space evaluation

 9.5 concluding remarks

 references

10 psvm: parallelizing support vector machines on distributed computers

 10.1 introduction

 10.2 interior point method with incomplete cholesky factorization

 10.3 psvm algorithm

10.3.1 parallel icf

10.3.2 parallel ipm

10.3.3 computing parameter b and writing back

 10.4 experiments

10.4.1 class-prediction accuracy

10.4.2 scalability

10.4.3 overheads

 10.5 concluding remarks

 references

11 approximate high-dimensional indexing with kernel

 11.1 introduction

 11.2 related reading

 11.3 algorithm spheredex

11.3.1 create - building the index

11.3.2 search - querying the index

11.3.3 update - insertion and deletion

 11.4 experiments

11.4.1 setup

11.4.2 performance with disk ios

11.4.3 choice of parameter g

11.4.4 impact of insertions

11.4.5 sequential vs. random

11.4.6 percentage of data processed

11.4.7 summary

 11.5 concluding remarks

11.5.1 range queries

11.5.2 farthest neighbor queries

 references

12 speeding up latent dirichlet allocation with parallelization and pipeline strategies

 12.1 introduction

 12.2 related reading

 12.3 ad-lda: approximate distributed lda

12.3.1 parallel gibbs sampling and allreduce

12.3.2 mpi implementation of ad-lda

 12.4 plda+

12.4.1 reduce bottleneck of ad-lda

12.4.2 framework of plda+

12.4.3 algorithm for pw processors

12.4.4 algorithm for pd processors

12.4.5 straggler handling

12.4.6 parameters and complexity

 12.5 experimental results

12.5.1 datasets and experiment environment

12.5.2 perplexity

12.5.3 speedups and scalability

 12.6 large-scale applications

12.6.1 mining social-network user latent behavior

12.6.2 question labeling (ql)

 12.7 concluding remarks

 references