Table of Contents

Chapter 1 Introduction
1.1 Static and Dynamic Artificial Intelligence
1.2 Theory of Continual Learning
1.2.1 Scenarios of Continual Learning
1.2.2 Challenges of Continual Learning
1.2.3 Approaches of Continual Learning
1.3 Content of This Book
Chapter 2 Multi-Domain Multi-Task Rehearsal for Continual Learning
2.1 Introduction
2.2 Methodology
2.2.1 Multi-Domain Multi-Task Rehearsal
2.2.2 Two-Level Angular Margin Loss
2.2.3 Episodic Distillation
2.2.4 Total Algorithm
2.3 Experiments
2.3.1 Experimental Settings
2.3.2 Comparison with the State-of-the-arts
2.3.3 Domain Shift Observation
2.4 Chapter Conclusion
Chapter 3 Exploring Example Influence in Continual Learning
3.1 Introduction
3.2 Methodology
3.2.1 Preliminary: Rehearsal-based CL
3.2.2 Example Influence on Stability and Plasticity
3.3 Meta Learning on Stability and Plasticity
3.3.1 Influence Function for SP
3.3.2 Simulating IF for SP
3.4 Using Influence for Continual Learning
3.4.1 Before Using: Influence for SP Pareto Optimality
3.4.2 Model Update Using Example Influence
3.4.3 Rehearsal Selection Using Example Influence
3.5 Experiments
3.5.1 Datasets and Implementation Details
3.5.2 Main Comparison Results
3.5.3 Analysis of Dataset Influence on SP
3.5.4 Analysis on SP Pareto Optimum
3.5.5 Training Time
3.6 Chapter Conclusion
Chapter 4 Measuring Asymmetric Gradient Discrepancy in Parallel Continual Learning
4.1 Introduction
4.2 Methodology
4.2.1 Parallel Continual Learning
4.2.2 Measuring Asymmetric Gradient Discrepancy
4.2.3 Maximum Discrepancy Optimization
4.3 Experiments
4.3.1 Dataset
4.3.2 Experiment Details
4.3.3 Main Results
4.3.4 Rehearsal Analysis in PCL
4.3.5 Comparison with Symmetric Metrics
4.3.6 Ablation Study
4.3.7 Procedure Time
4.3.8 Tolerance Analysis in AGD
4.4 Chapter Conclusion
Chapter 5 Multi-Label Continual Learning Using Augmented Graph Convolutional Network
5.1 Introduction
5.2 Methodology
5.2.1 Definition of MLCL
5.2.2 MLCL Scenarios
5.2.3 Overview of the Proposed Method
5.2.4 Partial Label Encoder
5.2.5 Augmented Correlation Matrix
5.2.6 Objective Function
5.3 Experiments
5.3.1 Datasets
5.3.2 Evaluation Metrics
5.3.3 Implementation Details
5.3.4 Baseline Methods
5.3.5 Main Results
5.3.6 More MLCL Settings
5.3.7 mAP Curves
5.3.8 Ablation Studies
5.3.9 Visualization of ACM
5.4 Chapter Conclusion
Chapter 6 Towards Long-Term Remembering for Federated Continual Learning
6.1 Introduction
6.1.1 Federated Learning
6.1.2 Federated Continual Learning
6.2 Methodology
6.2.1 Problem Definition
6.2.2 Multi-Node Collaborative Integration for Parameter Co-Importance
6.2.3 Fisher Accumulating and Balancing for Reducing Forgetting
6.3 Experiments
6.3.1 Experiment Details
6.3.2 Results
6.3.3 Ablation Experiment
6.3.4 Fisher Visualization
6.4 Chapter Conclusion
Chapter 7 Centroid-based Rehearsal in Online Continual Learning
7.1 Introduction
7.2 Methodology
7.2.1 Continual Domain Shift in OCL
7.2.2 Centroid-based Rehearsal
7.2.3 Distillation on Centroid Distance
7.2.4 The Overall Algorithm
7.3 Experiments
7.3.1 Dataset and Experimental Details
7.3.2 Experimental Results
7.4 Chapter Conclusion
Chapter 8 Dynamic V2X Perception from Road-to-Vehicle Vision
8.1 Introduction
8.2 Methodology
8.2.1 Overview
8.2.2 Overcoming Intra-Scene Changes
8.2.3 Overcoming Inter-Scene Changes
8.2.4 The Whole Algorithm
8.2.5 Bandwidth Discussion
8.3 Experiments
8.3.1 Data Preparation
8.3.2 Evaluation Metric
8.3.3 Compared Methods
8.3.4...