Machine learning and its application 2013 content introduction Books
Computer Books
Machine learning is a very important area of research in computer science and artificial intelligence. In recent years, machine learning has not only been a great skill in many fields of computer science, but also an important supporting technology for interdisciplinary disciplines. This book invites experts from the relevant fields at home and abroad to write a summary of the different branches of machine learning and related fields of research progress. The book is divided into 8 chapters, which are related to sparse topic learning, manifold learning and sequencing based on vector field, rank minimization, real value multi-variable dimension reduction and other applications such as knowledge mining and user modeling, heterogeneous face image synthesis, and the utility of multi-views in the use of unlabeled data in learning. A discussion of generalized correlation analysis for Applied view data.
"Machine learning and its application 2013" for universities, research institutes, computer, automation and related professional teachers and students, science and technology workers and related enterprises engineering and technical personnel to read reference. Catalogue "machine learning and its application 2013"
Learning Sparse Topical Representationsjun Zhuaonan zhangeric P. Xing
1 Introduction
2 related work
2.1Probabilistic LDA
2.2non-negative Matrix Factorization
3 Sparse Topical Coding
3.1A Probabilistic generative Process
3.2STC for MAP estimation
3.3Optimization with coordinate descent
4 Extensions
4.1Collapsed STC
4.2Supervised Sparse Topical Coding
5 experiments
5.1Sparse Word Code
5.2Prediction Accuracy
5.3Time efficiency
6 conclusion
References
The utility Wang Wei of multi-views in the use of unlabeled data learning Zhou Zhihua
1 Introduction
The utility of 2 multi-views in semi-supervised learning
The utility of 3 multi-views in active learning
The utility of 4 multi-views in active semi-supervised learning
5 View Split
6 concluding remarks
Reference documents
Knowledge Mining and user modeling Wang Haifeng Shiqi Zhao to Wei Reprieve Hao Wu Chang
1 Introduction
2 Technical Overview
3 Ontology Knowledge System construction
3.1 Knowledge Mining
3.2 Knowledge Processing
3.3 Semantic Computing
3.4 Experimental Results
3.5 Construction of the subject system based on ontology knowledge
4 cross-product user log Mining
4.1 Technical Framework
4.2 Cross-Product user Data Session Segmentation
4.3 Cross-product user data point of focus mining
5 User Modeling
5.1 User Attribute Modeling
5.2 User Interest Modeling
5.3 User State Modeling
5.4 Multi-dimensional user behavior analysis model
Regional correlation analysis of 5.5 user interest model
6 conclusion
Reference documents
Heterogeneous human face graphs for the synthesis of high-tech wave Wang Nanan
1 Introduction
2 Method of image synthesis based on subspace learning
2.1 Method based on linear subspace learning
2.2 A method based on manifold learning
3 Synthesis method based on Bayesian inference
3.1 Method based on embedded-human hidden Markov model
3.2 Method based on Markov random field
4 Synthesis method based on the thought of human face illusion
5 Experimental results
6 concluding remarks
Reference documents
Generalized correlation Analysis Sage AP Chen Songzan for Applied View data
1 Introduction
1.1 Multi-View data
The significance and method of 1.2 data dimensionality reduction
2 problems and solutions of dimensionality reduction method based on correlation analysis
2.1 Oversight information for ignoring multi-view data
2.2 requires full pairing of data between different views
2.3 Existing Solutions
3 Our research work
3.1 Semi-paired local correlation analysis
3.2 Semi-supervised semi-paired generalized correlation analysis
3.3 Neighborhood Correlation Analysis
4 Summary
Reference documents
He Xiaofei of manifold learning and sequencing based on vector field
1 Introduction
2 parallel vector fields and linear functions
2.1 Semi-supervised learning problems in manifolds
2.2 Parallel vector fields and linear functions
2.3 Objective function
3 discretization and optimization
3.1 Tangent space and vector field discretization
3.2 Gradient Field calculation
3.3 Parallel vector field calculation
3.4 Objective functions in discrete form
3.5 Objective Function Optimization
4 ordering based on the regularization of the parallel vector field
4.1 Regularization of the vector field
Discretization of 4.2 R1 and R2
4.3 Discretization of objective functions
4.4 Objective Function Optimization
4.5 Experiments
5 conclusion and Prospect
Reference documents
Rank minimization: Theory, algorithm and application Lin Zuechen
1 Introduction
2 Main mathematical models
3 Theoretical analysis
4 algorithm
4.1 Accelerated nearest Neighbor gradient method and its generalization
4.2 interleaving Direction method and its linearization
4.3 Calculation of singular value decomposition
5 Applications
5.1 Background Modeling
5.2 Image Batch Alignment
5.3 Transform invariant low rank texture
5.4 Motion Segmentation
5.5 Image Segmentation
5.6 Significant area detection of images
6 concluding remarks
Reference documents
Real-valued multivariate dimensionality of about simple Hongming Hawaii
1 Introduction
2 Real-valued multivariate dimensionality reduction
2.1-Slice Inverse regression method
Extension of inverse regression of 2.2 slices
2.3 Main Hessian Direction
2.4 Introduction to sub-spaces
2.5 Sparse full dimensionality reduction
2.6 Nuclear dimensionality reduction
2.7 Reduction of the minimum square dimension
3 reduction of the kernel dimension of tree-shaped structure
3.1 Motive
3.2 Introduction to the tree-shaped algorithm
3.3 (residual) tree core dimensionality reduction
3.4 Experimental Section
3.5 Conclusion
Application of 4 nuclear dimensionality reduction in population counting
4.1 Nuclear dimensionality reduction
4.2 Multi-core learning
5 Conclusion
Reference machine learning and its application 2015
- Gao Nio, Zhang Junping
- Series: Academic books of the Chinese Society of Computer Science--Knowledge sciences
- Publishing house: Tsinghua University Press
- ISBN:9787302406594
- Last Date: 2015-10-16
- Publication date: October 2015
- Folio: 16 Open
- Page: 228
- Edition: 1-1
- Category: Computer > Artificial Intelligence > Synthesis
Content Introduction Books
Computer Books
This book is a summary of the 11th and 12 session of China machine learning and its application seminar, and invited 10 experts from the Conference to write about their research fields, and discussed the research progress in different branches and related fields of machine learning in the form of a review. The book is divided into 10 chapters, which are related to sparse learning, implicit category analysis in crowdsourcing data, evolutionary optimization, deep learning, semi-supervised support vector machines, differential privacy protection, and machine learning applications in image quality evaluation, Image semantic segmentation, multi-modal image analysis, etc. The research progress of the new Hardware Cambrian Neural network computer is also introduced.
This book can be used for reference by researchers, teachers, postgraduates and engineers of computer, automation and related majors. The application of directory sparse learning in multi-task learning Yuan Pinghua Zhang Changshui
1 Introduction
2 robust multi-tasking feature learning
3 Multi-stage multi-task feature learning
4 Conclusion
Reference documents
Implicit category analysis in crowdsourcing data labeling Towada Zhu June
1 Introduction
2 Crowdsourcing labeling issues
3 several basic models of labeling integration
3.1 Multiple voting models
3.2 Confusion Matrix model
4 implicit category structure in crowdsourcing annotations
5 Implicit category Estimation
6 Experimental performance
7 Conclusion
Reference documents
Advances in theoretical research of evolutionary optimization Youyan
1 Introduction
2 Evolutionary optimization algorithm
3 Theoretical development of evolutionary optimization
4 Run Time Analysis method
5 Approximation Performance Analysis
6 Algorithm parameter Analysis
7 Conclusion
Reference documents
Deep learning algorithm based on Bayesian convolution network Chen Bo
1 Introduction
2 multi-layer sparse factor analysis
2.1 Single-layer model
2.2 Decimation and maximum pooling
2.3 Model features and visualizations
3 Hierarchical Bayesian analysis
3.1 Hierarchical structure
3.2 Calculation
3.3 Application of Bayesian output
3.4 Relevance to the previous model
4 exploration of convolution in inference
4.1Gibbs Sampling
4.2VB Inference
4.3 online vb
5 Experimental results
5.1 Parameter setting
5.2 Synthetic data and mnist data
5.3Caltech 101 Data Analysis
5.4 Activation of each layer
5.5 sparsity
5.6 Classification for the Caltech 101
5.7 Online VB and van Gogh oil painting analysis
6 conclusion
Reference documents
Study on the learning method of semi-supervised support vector machine Li Yu Zhou Zhihua
1 Introduction
Introduction to 2 semi-supervised support vector machines
3 semi-supervised support vector machine learning method
More than 3.1: large-scale semi-supervised support vector machines for multi-training examples
3.2 Fast: Fast semi-supervised support vector machine for improving learning efficiency
3.3 Good: Secure semi-supervised support vector machine for performance assurance
3.4 Provinces: Cost-sensitive semi-supervised support vector machines for cost suppression
4 Conclusion
Reference documents
Machine learning Wang Liweizhenke for differential privacy protection
1 Introduction
2 Related definitions and properties
3 Common mechanisms
4 privacy protection mechanism for smooth queries
5 Experimental results
6 conclusion
Reference documents
Study on the evaluation method of non-reference image quality Gao Nio He Li Fire
1 Introduction
2 Method of image quality evaluation based on feature representation
2.1 Non-reference image quality evaluation method based on feature dimension reduction
2.2 Evaluation method of image quality without reference based on image block learning
2.3 Non-reference image quality evaluation method based on sparse representation
3 Method of image quality evaluation based on regression analysis
3.1 Evaluation method of non-reference image quality based on support vector regression
3.2 Evaluation method of non-reference image quality based on neural network
3.3 A non-reference image quality evaluation method based on multi-core learning
4 Method of image quality evaluation based on Bayesian inference
4.1 Simple probabilistic model image quality evaluation method
4.2 Method of image quality evaluation based on subject probability model
4.3 Method of image quality evaluation based on deep learning
5 Experimental results
6 conclusion
Reference documents
Image Semantic Segmentation Shi Shangyang
1 Introduction
2 Unsupervised Image Region segmentation
3 The method of full supervised semantic segmentation
3.1 Semantic Segmentation method based on multi-scale segmentation
3.2 Semantic Segmentation method based on multi-feature fusion
3.3 Semantic Segmentation method based on depth network
4 The semantic segmentation method of weak supervision
4.1 Training image data with bounding box
4.2 Training image data with accurate image layer labels
4.3 Training image data with noise tag
5 Common data sets for semantic image segmentation
6 Comparison of state of the art method under different supervisory conditions
7 Conclusion
Reference documents
Application of machine learning in multi-modal brain image analysis Zhang Daojiang waveforms Liu Mingxia
1 Introduction
2 manifold regularization multi-task feature learning
3 multi-modal manifold regularization migration learning
4 views centralized multi-spectral classification
5 Experimental results
5.1 Manifold regularization multi-task feature learning
5.2 Multi-modal manifold regularization migration learning
5.3 Views centralized multi-spectral classification
6 conclusion
Reference documents
Cambrian Neural network computer Chen Tianxi Chen Yun Ji
1 Artificial Neural network
2 Past Failures
2.1 Algorithm: The Rise of SVM
2.2 Application: Cognitive tasks are ignored
2.3 Process: Universal processors enjoy Moore's Law dividend
3 The Nirvana of the Neural network computer
3.1 algorithm: An effective training algorithm for deep learning
3.2 Application: Universalization of cognitive tasks
3.3 Technology: The advent of the dark silicon era
3.4 The rise of the second generation of neural networks
4 major challenges
5 Cambrian Neural Network (machine learning) processor
5.1DianNao
5.2DaDianNao
5.3PuDianNao
6 Future work
References ↑ Folding Preface with the advent of the era of big data, data from the Internet, security, finance, medical, scientific observation and many other fields are exploding. While enjoying the wealth of information provided by huge amounts of data, we are also drowning in the ocean of data, and it is difficult to dig out much-needed information and the most useful knowledge. To solve this contradiction, an important strategy is to take advantage of machine learning.
Machine learning originates from artificial intelligence, which has gradually developed into a relatively complete and independent subject in recent 30 years, and is widely concerned by computer science, statistics, cognitive science and other related fields. In theory, according to the difference between data sampling distribution and real distribution, the learning mechanism of probability approximation approximation (PAC) is formed, and the traditional statistical learning theory is developed on this basis. In order to avoid the ill-posed problem of objective function in data prediction, a series of regularization theories are proposed, such as the sparse learning technique focusing on explanatory, the manifold regularization theory, which is focused on preserving the nonlinear geometric structure of data, and the maximum interval regularization technique which is expected to maintain the optimal classification performance. In addition, application-driven machine learning also drives a number of emerging research directions, such as semi-supervised learning without tagged data, migration learning with different data distributions, Domain adaptive learning, multi-label, multi-sample, multi-view, multi-task learning with Data "many" features, Consider the network data tagging strategy of crowdsourcing learning and so on. The fusion of optimization techniques such as stochastic gradient descent and semi-positive definite planning has also facilitated the processing of large-scale data and the solution of global optimization. It is worth mentioning that in recent years, depth (neural network) learning through the progressive reduction of feature extraction technology and big data training strategy, in many layers of image, voice and even text classification performance beyond the statistical learning-oriented machine learning method. This makes the neural network after nearly more than 20 years of trough, once again the vast number of researchers to re-attract the eyeball back. It not only set off a new wave of machine learning, but also directly led to the industry's study of machine learning and development of unprecedented attention.
In 2002, academician Mr Peter Luk Qian organized the "Intelligent Information Processing Series seminar" at Fudan University's Intelligent Information Processing laboratory and listed "machine learning and its application" as one of the workshops supported by the year. In November 2002, the seminar was held successfully and the purpose of the Conference was determined, such as non-essay, no charge, invitation from organizer, and "academic supremacy, other simple". In November 2004, the second seminar on "Machine learning and its application" was held at Fudan University, with more than 100 people attending two and a half days of meetings. Since 2005, the symposium was organized by the State Key laboratory of software technology at Nanjing University. The third seminar, held in November 2005, attracted more than 250 people from nearly 10 provinces and cities nationwide, and in November 2006 and November 2007, the fourth and fifth sessions were organized by the School of Information Science and Technology of Nanjing University of Aeronautics and Astronautics and the School of Mathematics and Computer sciences Two times attracted about 300 people from more than 10 provinces and municipalities in the country, the sixth session held in November 2008 coincided with the 50 anniversary of the establishment of computer science in Nanjing University, attracting more than 10 people from more than 380 provinces and cities nationwide. Thereafter, the seventh to eighth session of the Seminar was held in November 2009 and November 2010 at Nanjing University, with about 400 people attending. In November 2011 and November 2012, the Nineth and tenth sessions were held by Tsinghua University's state Key laboratory of automation, intelligent science and systems, and the National Laboratory for Information Science and Technology of Tsinghua University, where more than 500 people were attending the two meetings. The 11th session was held in November 2013 by the Fudan University Computer Science Technology Institute and the Shanghai Intelligent Information Processing Laboratory, and the 12th session was held in Xidian City in November 2014, with more than 600 participants attending the meeting. It can be said that the "machine learning and its Application" seminar has become a gathering of researchers in machine learning and related fields.
This book is a summary of the 11th and 12 session of China machine learning and its application seminar, and invited 10 experts from the Conference to write about their research fields, and discussed the research progress in different branches and related fields of machine learning in the form of a review. The book is divided into 10 chapters, which are related to sparse learning, implicit category analysis in crowdsourcing data, evolutionary optimization, deep learning, semi-supervised support vector machines, differential privacy protection, and machine learning applications in image quality evaluation, Image semantic segmentation, multi-modal image analysis, etc. The research progress of the new Hardware Cambrian Neural network computer is also introduced.
Among them, Dr. Yuan Pinghua and Professor Zhang Changshui in the 1th chapter studied the theory and algorithm of sparse learning in robust multi-task feature learning and multi-stage multi-task feature learning. In the 2nd Chapter, Dr. Towada and Professor Zhu June summarized the two basic models of crowdsourcing labeling and labeling integration, and proposed a model of implicit category structure in crowdsourcing learning. In the 3rd chapter, Professor Youyan analyzes the theoretical basis of evolutionary optimization for most evolutionary algorithms, which often depend on the insufficiency of heuristic algorithms. By drawing on the multi-layered framework of deep learning, Professor Chen Yu has developed hierarchical Bayesian analysis and online variable decibel Dean inference method in the 4th chapter. In the 5th chapter, Dr. Li Yu and Professor Zhou Zhihua discussed and analyzed the new progress of semi-supervised support vector machine in the last ten years from the four aspects of "many", "fast", "good" and "province". Considering that most machine learning algorithms are based on data sets containing user-sensitive information, Professor Liwei Wang and Dr Zheng Kai in the 6th chapter analyzes the advantages and disadvantages of the existing privacy protection model, and proposes a privacy protection mechanism for smooth queries based on the differential privacy policy. As one of the most important vectors in visual data, the quality evaluation of image is the basic problem of visual information quality evaluation. In the 7th chapter, Professor Gao Nio and Dr Heli introduced a method of machine learning based on feature representation, regression analysis and Bayesian inference to evaluate the image quality objectively, and put forward a series of measures to evaluate the image quality without reference. In addition, considering that the high-level semantic extraction of images has always been a "difficult" problem in computer vision, Professor Shi Shangyang in the 8th chapter analyzes the semantic segmentation of images from the aspects of feature fusion, depth network and weak supervisory strategy. In brain image analysis, multiple acquisition devices can form multi-modal images. In order to effectively fuse multi-modal heterogeneous brain image data, Professor Zhang Daojiang and Dr. Waveforms in the 9th chapter, the application of machine learning in multi-modal brain image analysis is analyzed from multi-task learning, multi-modal manifold transfer learning and multi-view classification. Finally, Chen Tianxi and Professor Chenyun discussed the possibility of machine learning curing from the hardware point of view, and gave a brief introduction to the Cambrian series of processors developed by the Institute of Computing Technology of CAS.
This book summarizes the latest research progress of machine learning and its application in China, which can be used as a reference for researchers, teachers, postgraduates and engineers of computer, automation, information processing and related majors, as well as auxiliary content of artificial intelligence and machine learning courses, hoping to help people who are interested in machine learning research.
Gao Nio Zhang Junping
July 2015
Machine learning and its Applications 2013, machine learning and its applications
