Information Fusion: Machine Learning Methods

✍ Scribed by Jinxing Li, Bob Zhang, David Zhang

• Publisher: Springer
• Year: 2022
• Tongue: English
• Leaves: 283
• Category: Library

✦ Synopsis

In the big data era, increasing information can be extracted from the same source object or scene. For instance, a person can be verified based on their fingerprint, palm print, or iris information, and a given image can be represented by various types of features, including its texture, color, shape, etc. These multiple types of data extracted from a single object are called multi-view, multi-modal or multi-feature data. Many works have demonstrated that the utilization of all available information at multiple abstraction levels (measurements, features, decisions) helps to obtain more complex, reliable and accurate information and to maximize performance in a range of applications.

This book provides an overview of information fusion technologies, state-of-the-art techniques and their applications. It covers a variety of essential information fusion methods based on different techniques, including sparse/collaborative representation, kernel strategy,Bayesian models, metric learning, weight/classifier methods, and deep learning. The typical applications of these proposed fusion approaches are also presented, including image classification, domain adaptation, disease detection, image restoration, etc.

This book will benefit all researchers, professionals and graduate students in the fields of computer vision, pattern recognition, biometrics applications, etc. Furthermore, it offers a valuable resource for interdisciplinary research.

✦ Table of Contents

Preface
Acknowledgments
Contents
List of Figures
List of Tables
1 Introduction
1.1 Why Do Information Fusion?
1.2 Related Works
1.2.1 Multi-View Based Fusion Methods
1.2.2 Multi-Technique Based Fusion Methods
1.3 Book Overview
References
2 Information Fusion Based on Sparse/Collaborative Representation
2.1 Motivation and Preliminary
2.1.1 Motivation
2.1.2 Preliminary
2.1.2.1 Sparse Representation Classifier
2.1.2.2 Collaborative Representation Classifier
2.2 Joint Similar and Specific Learning
2.2.1 Problem Formulation
2.2.2 Optimization for JSSL
2.2.3 The Classification Rule for JSSL
2.2.4 Experimental Results
2.2.4.1 Image Dataset
2.2.4.2 Healthy Versus DM Classification
2.2.4.3 Healthy Versus IGR Classification
2.2.5 Conclusion
2.3 Relaxed Collaborative Representation
2.3.1 Problem Formulation
2.3.2 Optimization for RCR
2.3.3 The Classification Rule for RCR
2.3.4 Experimental Results
2.3.4.1 FR in Controlled Environment
2.3.4.2 FR in Uncontrolled Environment
2.3.4.3 Object Categorization
2.3.5 Conclusion
2.4 Joint Discriminative and Collaborative Representation
2.4.1 Problem Formulation
2.4.2 Optimization for JDCR
2.4.3 The Classification Rule for JDCR
2.4.4 Experimental Results
2.4.4.1 Image Dataset
2.4.4.2 Experiments Set
2.4.4.3 Healthy Versus Fatty Liver Classification
2.4.4.4 Time Consumption
2.4.5 Conclusion
References
3 Information Fusion Based on Gaussian Process Latent Variable Model
3.1 Motivation and Preliminary
3.1.1 Motivation
3.1.2 Preliminary
3.1.2.1 GPLVM
3.1.2.2 SGPLVM
3.2 Shared Auto-encoder Gaussian Process Latent Variable Model
3.2.1 Problem Formulation
3.2.2 Optimization for SAGP
3.2.3 Inference
3.2.4 Experimental Results
3.2.4.1 Dataset Description
3.2.4.2 Experimental Settings
3.2.4.3 Performance on the Three Datasets
3.2.5 Conclusion
3.3 Multi-Kernel Shared Gaussian Process Latent Variable Model
3.3.1 Problem Formulation
3.3.1.1 Decoding Part
3.3.1.2 Priors for Classification
3.3.1.3 Encoding Part
3.3.2 Optimization for MKSGP
3.3.3 Inference
3.3.4 Experimental Results
3.3.4.1 Experimental Settings
3.3.4.2 Experimental Results
3.3.5 Conclusion
3.4 Shared Linear Encoder-Based Multi-Kernel Gaussian Process Latent Variable Model
3.4.1 Problem Formulation
3.4.2 Optimization for SLEMKGP
3.4.3 Inference
3.4.4 Experimental Results
3.4.4.1 Experimental Setting
3.4.4.2 Result Comparison
3.4.5 Conclusion
References
4 Information Fusion Based on Multi-View and Multi-Feature Learning
4.1 Motivation
4.2 Generative Multi-View and Multi-Feature Learning
4.2.1 Problem Formulation
4.2.2 Optimization for MVMFL
4.2.3 Inference for MVMFL
4.2.4 Experimental Results
4.2.4.1 Datasets and Experimental Setting
4.2.4.2 Experimental Results on Synthetic Data
4.2.4.3 Experimental Results on Biomedical Data
4.2.5 Conclusion
4.3 Hierarchical Multi-View Multi-Feature Fusion
4.3.1 Problem Formulation
4.3.2 Optimization for HMMF
4.3.3 Inference for HMMF
4.3.4 Experimental Results
4.3.4.1 Datasets and Experimental Setting
4.3.4.2 Experimental Results on Datasets
4.3.5 Conclusion
References
5 Information Fusion Based on Metric Learning
5.1 Motivation
5.2 Generalized Metric Swarm Learning
5.2.1 Problem Formulation
5.2.2 Optimization for GMSL
5.2.3 Solving with Model Modification
5.2.4 Representation of Pairwise Samples in Metric Swarm Space
5.2.5 Sample Pair Verification
5.2.6 Remarks
5.2.6.1 Joint Metric Swarm Score Function
5.2.6.2 Optimality Condition
5.2.7 Experimental Results
5.2.7.1 Parameter Setting
5.2.7.2 Test Results on UCI Datasets
5.2.7.3 Test Results on LFW Faces
5.2.7.4 Test Results on PubFig Faces
5.2.8 Conclusion
5.3 Combined Distance and Similarity Measure
5.3.1 Problem Formulation
5.3.1.1 CDSM and Pairwise Kernel Explanation
5.3.1.2 Triplet-Based Learning Model
5.3.2 Optimization for CDSM
5.3.3 Kernelized CDSM
5.3.4 Experimental Results
5.3.4.1 Evaluation on CDSM and Kernelized CDSM
5.3.4.2 Comparison with the State-of-the-Art Metric Learning Methods
5.3.4.3 Results on the UCI Datasets
5.3.4.4 Results on the Handwritten Digit Datasets
5.3.4.5 CDSM VS. Kernelized CDSM
5.3.4.6 Comparison of Running Time
5.3.5 Conclusion
References
6 Information Fusion Based on Score/Weight Classifier Fusion
6.1 Motivation
6.2 Adaptive Weighted Fusion Approach
6.2.1 Problem Formulation
6.2.2 Rationale and Advantages of AWFA
6.2.3 Experimental Results
6.2.3.1 Experiments on the HFB
6.2.3.2 Experiments on the 2D Plus 3D Palmprint Dataset
6.2.3.3 Experiments on the PolyU Multispectral Palmprint Dataset
6.2.3.4 Experiments on the Georgia Tech Face Dataset
6.2.3.5 Experiments on the LFW Dataset
6.2.4 Conclusions
6.3 Adaptive Weighted Fusion of Local Kernel Classifiers
6.3.1 FaLK-SVM
6.3.2 Adaptive Fusion of Local SVM Classifiers
6.3.2.1 The Adaptive Fusion Method
6.3.2.2 Model Selection
6.3.3 Experimental Results
6.3.3.1 Experimental Results on the UCI Datasets
6.3.3.2 Experimental Results on the Large-Scale Datasets
6.3.4 Conclusion
References
7 Information Fusion Based on Deep Learning
7.1 Motivation
7.2 Dual Asymmetric Deep Hashing Learning
7.2.1 Problem Formulation
7.2.2 Optimization for DADH
7.2.3 Inference for DADH
7.2.4 Experimental Results
7.2.4.1 Datasets
7.2.4.2 Baseline and Evaluation Protocol
7.2.4.3 Implementation
7.2.4.4 Comparison with Other Methods
7.2.4.5 Convergence Analysis
7.2.5 Conclusion
7.3 Relaxed Asymmetric Deep Hashing Learning
7.3.1 Problem Formulation
7.3.2 Optimization for RADH
7.3.3 Inference for RADH
7.3.4 Implementation
7.3.5 Experimental Results
7.3.5.1 Experimental Settings
7.3.5.2 Comparison with Other Methods
7.3.6 Conclusion
7.4 Joint Learning of Single-Image and Cross-Image Representations for Person Re-identification
7.4.1 Joint SIR and CIR Learning
7.4.1.1 Connection Between SIR and CIR
7.4.1.2 Pairwise Comparison Formulation
7.4.1.3 Triplet Comparison Formulation
7.4.1.4 Prediction
7.4.2 Deep Convolutional Neural Network
7.4.2.1 Network Architecture
7.4.2.2 Network Training
7.4.3 Experiments
7.4.3.1 CUHK03 Dataset
7.4.3.2 CUHK01 Dataset
7.4.3.3 VIPeR Dataset
7.4.4 Conclusion
References
8 Conclusion
Index