Reproduce “Online Continual Learning: A Systematic Literature Review of Approaches, Challenges, and Benchmarks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Seyed Amir Bidaki, Amir Mohammadkhah, Kiyan Rezaee, Faeze Hassani, Sadegh Eskandari, Maziar Salahi, Mohammad M. Ghassemi
A systematic literature review synthesized findings from 81 unique Online Continual Learning (OCL) approaches, analyzing over 1,000 features and identifying more than 500 components. This comprehensive overview provides a structured understanding of the current state of OCL research, detailing common strategies, architectures, datasets, and key challenges like catastrophic forgetting.
Key insights
- Replay-based methods are the most prevalent strategy in Online Continual Learning (OCL), with Experience Replay being commonly used.
- ResNet architectures are the most popular backbone (used in 61 of 81 surveyed approaches), and SGD is the dominant optimization technique.
- Catastrophic forgetting remains the primary challenge in OCL, explicitly addressed by 44 of the analyzed approaches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.