Recent Advancements in Multi-View Data Analytics (Studies in Big Data, 106)

✍ Scribed by Witold Pedrycz (editor), Shyi-Ming Chen (editor)

• Publisher: Springer
• Year: 2022
• Tongue: English
• Leaves: 346
• Category: Library

✦ Synopsis

This book provides timely studies on multi-view facets of data analytics by covering recent trends in processing and reasoning about data originating from an array of local sources. A multi-view nature of data analytics is encountered when working with a variety of real-world scenarios including clustering, consensus building in decision processes, computer vision, knowledge representation, big data, data streaming, among others.

The chapters demonstrate recent pursuits in the methodology, theory, advanced algorithms, and applications of multi-view data analytics and bring new perspectives of data interpretation. The timely book will appeal to a broad readership including both researchers and practitioners interested in gaining exposure to the rapidly growing trend of multi-view data analytics and intelligent systems.

✦ Table of Contents

Preface
Contents
The Psychology of Conflictive Uncertainty
1 Introduction
2 Conflictive Uncertainty
3 Conflictive Uncertainty Aversion
4 Consequences of Conflictive Uncertainty for Risk Communication
5 Dealing with and Communicating About Conflictive Uncertainty
6 Conclusion and Suggestions for Multi-View Modeling Practices
References
How Multi-view Techniques Can Help in Processing Uncertainty
1 Introduction
2 Need for Uncertainty Quantification
3 What Makes Uncertainty Quantification Easier and What Makes It More Complex
4 How Uncertainty Quantification Is Related to Multi-view Techniques
5 Straightforward (``Naive'') Approach and Its Limitations
6 Monte-Carlo Approach: Traditional Probabilistic Case
7 Algorithm for the Probabilistic Case
8 Need to Go Beyond the Traditional Probabilistic Case
9 Case When We Know the Probability Distributions but They Are Not Necessarily Normal
10 Case of Interval Uncertainty
11 Final Algorithm for the Interval Case
12 What if We Have Information About Systematic and Random Error Components
13 Fuzzy Case
14 General Case: What if We Know Different Inputs with Different Uncertainty
References
Multi-view Clustering and Multi-view Models
1 Introduction
1.1 Clustering
1.2 Multi-view Clustering
2 Related Works
2.1 Some Related Concepts
2.2 Traditional Clustering Algorithms
2.3 Multi-view Data Sets
2.4 Validation Indexes
3 Multi-view Clustering Algorithms and Models
3.1 Simultaneous Weighting on Views and Features Algorithm
3.2 Collaborative Feature-Weighted Multi-view Fuzzy c-Means Clustering Algorithm
3.3 Multi-view Fuzzy Co-Clustering Algorithm
3.4 Fuzzy Optimization Multi-objective Clustering Ensemble Model for Multi-source Data Analysis
4 Some Research Directions on Multi-view Data
4.1 Information Search Problem
4.2 The Problem of Ground Observation Data Analyzing
4.3 The Problem of System Management and Operation
4.4 The Problem of Predicting the Status
5 Conclusion and Future Works
References
Rethinking Collaborative Clustering: A Practical and Theoretical Study Within the Realm of Multi-view Clustering
1 Introduction
2 Collaborative Clustering: State of the Art of a Polymorphic Notion with Very Diverse Applications
2.1 Evolution of the Notion of Collaborative Clustering
2.2 Remarkable Branches of Collaborative Clustering and Applications that Blur the Lines Between Ensemble Learning and Multi-view Clustering
3 Distinguishing Regular Clustering, Collaborative Clustering, Multi-view Clustering and Unsupervised Ensemble Learning
3.1 Notations
3.2 Definitions, Context, and Practical Setting
3.3 Summary: Four Interleaving Notions
4 Properties of Collaborative Clustering: Stability, Novelty and Consistency
4.1 Reminders on Clustering Stability
4.2 From Regular to Collaborative Clustering: Stability, Novelty and Consistency
4.3 Stability of Collaborative Clustering
5 Open Questions
6 Conclusion
References
An Optimal Transport Framework for Collaborative Multi-view Clustering
1 Introduction
2 Related Works
3 Optimal Transport
3.1 Monge's Formulation
3.2 Kantorovich's Relaxation
3.3 Discrete Settings
3.4 Wasserstein Distance
3.5 Wasserstein Barycenter
3.6 Entropy-Regularization
4 Proposed Approaches
4.1 Motivation
4.2 Collaborative Multi-view Clustering
5 Experimental Validation
5.1 Setting
5.2 Results and Validation
6 Conclusion
References
Data Anonymization Through Multi-modular Clustering
1 Introduction
2 Background Theory
2.1 State of the Art: Existed Approaches for Achieving k-Anonymity via Microaggregation
2.2 Multi-view Unsupervised Collaborative Clustering
3 Proposed Methods of Anonymization
3.1 Pre-Anonymization Step
3.2 Fine Tuning
3.3 Integrating Discriminatory Power
4 Validation of the Approaches
4.1 Description of the Datasets
4.2 Utility Measures and Statistical Analysis
5 Conclusion
References
Multi-view Clustering Based on Non-negative Matrix Factorization
1 Introduction
2 Related Works
2.1 Topological Collaborative Clustering
2.2 Classical NMF
2.3 Hard K-Means and NMF
3 Multi-view Setting
3.1 Optimized Weights for the Collaborative NMF to Improve Stability
4 Experimental Results
4.1 Purity Validation Method
4.2 Silhouette Validation Method
4.3 Datasets
4.4 Illustration of the Method on the Waveform Dataset
4.5 Validation Using Additional Datasets
5 Conclusions
6 Appendix
6.1 Gradients Computations
References
A Graph-Based Multi-view Clustering Approach for Continuous Pattern Mining
1 Introduction
2 Related Work
2.1 Multi-view Clustering Algorithms
2.2 Stream Clustering Algorithms
2.3 Multi-view Stream Clustering Algorithms
3 Background
3.1 Minimum Spanning Tree Clustering
3.2 Non-negative Matrix Factorization
3.3 Cluster Validation Measures
4 MST-MVS Clustering Algorithm
4.1 Multi-view Data Integration
4.2 Extraction of Multi-view Patterns
4.3 Transfer of Knowledge Through Artificial Nodes
4.4 CNMF-Based Labelling Algorithm
4.5 Pattern-Based Labelling Algorithm
4.6 Computational Complexity
5 Data and Experimental Settings
5.1 Data
5.2 Data Preparation
5.3 Experiments and Validation
5.4 Implementation and Availability
6 Results and Discussion
6.1 Algorithm Configuration
6.2 Tuning of Algorithm Parameters
6.3 Evaluation of Algorithm Performance
7 Conclusion and Future Work
References
Learning Shared and Discriminative Information from Multiview Data
1 Introduction
2 Joint Knowledge Discovery from Multiview Data
2.1 Canonical Correlation Analysis
2.2 Graph-Regularized Multiview CCA
2.3 Generalization Bound of GMCCA
2.4 Graph-Regularized Kernel MCCA
2.5 Applications
3 Discriminative Knowledge Discovery from Multiview Data
3.1 (Contrastive) Principal Component Analysis Revisit
3.2 Discriminative Principal Component Analysis
3.3 Optimality of dPCA
3.4 Applications
4 Concluding Remarks
References
A Supervised Ensemble Subspace Learning Model Based on Multi-view Feature Fusion Employing Multi-template EMG Signals
1 Introduction
2 State of the Art of EMG Diagnosis
3 Focus of Our Algorithm
4 Methodology
4.1 Strategy I (S-I)
4.2 Strategy II (S-II)
4.3 Preprocessing, Signal Selection and Orientation
4.4 Multi-view Learning
4.5 Feature Extraction and Reduction
4.6 Feature Fusion, Transformation and Classification
4.7 Performance Evaluation Markers
5 Results and Discussion
5.1 Database Description
5.2 Performance of the mCCA on Real-Time Dataset
5.3 Reliability and Scalability of the mCCA
6 Comparative Analysis
7 Conclusion
References
Performance Profiling of Operating Modes via Multi-view Analysis Using Non-negative Matrix Factorisation
1 Introduction
2 Related Work
2.1 Profiling with NMF
2.2 Multi-view Data Analysis Approaches
2.3 Multi-view Data Analysis Techniques Based on NMF
3 Materials and Methods
3.1 Data and Computer Code
3.2 Endogeneous Versus Exogeneous Data Views
3.3 Endogeneous Data View: Operating Mode Identification
3.4 Exogeneous Data View: Performance Profiling
3.5 Linking the Data Views: Operating Modes Versus Performance Profiles
4 Results
4.1 Identification of Endogeneous and Exogeneous Data Views
4.2 Endogeneous Data View: Operating Mode Identification
4.3 Exogeneous Data View: Performance Profiling
5 Conclusion
References
A Methodology Review on Multi-view Pedestrian Detection
1 Introduction
2 Monocular Pedestrian Detection
3 Low-Level Information Fusion
4 Intermediate-Level Information Fusion
5 High-Level Information Fusion
5.1 The Bottom-Up Approach
5.2 The Top-Down Approach
5.3 Pros and Cons of both Approaches
6 Deep Learning Based Information Fusion
6.1 Deep Front-End Models
6.2 Deep End-to-End Models
6.3 The Challenges
7 Performance Evaluation
7.1 Multi-view Video Datasets
7.2 Performance Metrics
8 Conclusions
References
Index