Robust representation for data analytics: models and applications

Description

This book introduces the concepts and models of robust representation learning, and provides a set of solutions to deal with real-world data analytics tasks, such as clustering, classification, time series modeling, outlier detection, collaborative filtering, community detection, etc. Three types of robust feature representations are developed, which extend the understanding of graph, subspace, and dictionary. Leveraging the theory of low-rank and sparse modeling, the authors develop robust feature representations under various learning paradigms, including unsupervised learning, supervised learning, semi-supervised learning, multi-view learning, transfer learning, and deep learning. Robust Representations for Data Analytics covers a wide range of applications in the research fields of big data, human-centered computing, pattern recognition, digital marketing, web mining, and computer vision.  Read more... Abstract: This book introduces the concepts and models of robust representation learning, and provides a set of solutions to deal with real-world data analytics tasks, such as clustering, classification, time series modeling, outlier detection, collaborative filtering, community detection, etc. Three types of robust feature representations are developed, which extend the understanding of graph, subspace, and dictionary. Leveraging the theory of low-rank and sparse modeling, the authors develop robust feature representations under various learning paradigms, including unsupervised learning, supervised learning, semi-supervised learning, multi-view learning, transfer learning, and deep learning. Robust Representations for Data Analytics covers a wide range of applications in the research fields of big data, human-centered computing, pattern recognition, digital marketing, web mining, and computer vision Content: Preface Contents 1 Introduction 1.1 What Are Robust Data Representations? 1.2 Organization of the Book Part I Robust Representation Models 2 Fundamentals of Robust Representations 2.1 Representation Learning Models 2.1.1 Subspace Learning 2.1.2 Multi-view Subspace Learning 2.1.3 Dictionary Learning 2.2 Robust Representation Learning 2.2.1 Subspace Clustering 2.2.2 Low-Rank Modeling References 3 Robust Graph Construction 3.1 Overview 3.2 Existing Graph Construction Methods 3.2.1 Unbalanced Graphs and Balanced Graph 3.2.2 Sparse Representation Based Graphs. 3.2.3 Low-Rank Learning Based Graphs3.3 Low-Rank Coding Based Unbalanced Graph Construction 3.3.1 Motivation 3.3.2 Problem Formulation 3.3.3 Optimization 3.3.4 Complexity Analysis 3.3.5 Discussions 3.4 Low-Rank Coding Based Balanced Graph Construction 3.4.1 Motivation and Formulation 3.4.2 Optimization 3.5 Learning with Graphs 3.5.1 Graph Based Clustering 3.5.2 Transductive Semi-supervised Classification 3.5.3 Inductive Semi-supervised Classification 3.6 Experiments 3.6.1 Databases and Settings 3.6.2 Spectral Clustering with Graph. 3.6.3 Semi-supervised Classification with Graph3.6.4 Discussions 3.7 Summary References 4 Robust Subspace Learning 4.1 Overview 4.2 Supervised Regularization Based Robust Subspace (SRRS) 4.2.1 Problem Formulation 4.2.2 Theoretical Analysis 4.2.3 Optimization 4.2.3.1 Learn Subspace P on Fixed Low-Rank Representations 4.2.3.2 Learn Low-Rank Representations Z on Fixed Subspace 4.2.4 Algorithm and Discussions 4.3 Experiments 4.3.1 Object Recognition with Pixel Corruption 4.3.2 Face Recognition with Illumination and Pose Variation 4.3.3 Face Recognition with Occlusions. 4.3.4 Kinship Verification4.3.5 Discussions 4.4 Summary References 5 Robust Multi-view Subspace Learning 5.1 Overview 5.2 Problem Definition 5.3 Multi-view Discriminative Bilinear Projection (MDBP) 5.3.1 Motivation 5.3.2 Formulation of MDBP 5.3.2.1 Learning Shared Representations Across Views 5.3.2.2 Incorporating Discriminative Regularization 5.3.2.3 Modeling Temporal Smoothness 5.3.2.4 Objective Function 5.3.3 Optimization Algorithm 5.3.3.1 Time Complexity Analysis 5.3.4 Comparison with Existing Methods 5.4 Experiments 5.4.1 UCI Daily and Sports Activity Dataset. 5.4.1.1 Two-View Setting5.4.1.2 Baselines 5.4.1.3 Classification Scheme 5.4.1.4 Results 5.4.2 Multimodal Spoken Word Dataset 5.4.2.1 Three-View Setting 5.4.2.2 Results 5.4.3 Discussions 5.4.3.1 Parameter Sensitivity and Convergence 5.4.3.2 Experiments with Data Fusion and Feature Fusion 5.5 Summary References 6 Robust Dictionary Learning 6.1 Overview 6.2 Self-Taught Low-Rank (S-Low) Coding 6.2.1 Motivation 6.2.2 Problem Formulation 6.2.3 Optimization 6.2.4 Algorithm and Discussions 6.3 Learning with S-Low Coding 6.3.1 S-Low Clustering 6.3.2 S-Low Classification.