A Comprehensive Survey of Continual Learning: Theory, Method and Application

By Liyuan Wang et al

Published on Aug. 10, 2015

Read the original document by opening this link in a new tab.

Table of Contents

1. Abstract - To cope with real-world dynamics... 2. Introduction - Learning is the basis for intelligent systems... 3. Setup - Basic Formulation, Typical Scenario, Evaluation Metric... 4. Theoretical Foundation - Stability-Plasticity Trade-off... 5. Representative Methods - Regularization-based approach, Replay-based approach... 6. Realistic Applications - Scenario complexity, Task specificity... 7. Discussion - Current trends, Cross-directional prospects...

Summary

This document provides a comprehensive survey of continual learning, focusing on the theory, methods, and applications in the field. It addresses the challenges of balancing stability and plasticity in learning new tasks while retaining knowledge of old tasks. Various approaches, such as the replay-based method and Bayesian approximation, are discussed. The paper also delves into realistic applications and the adaptation of methods to different scenarios. Overall, it offers insights into the evolving landscape of continual learning.