15 Continual Learning & Online Adaptation
How can an intelligent system change without forgetting what made it capable?
Area 15: Continual Learning & Online Adaptation
Master Question
How can an intelligent system change without forgetting what made it capable?
What We Want to Discover
Learning a new task or adapting to a new condition often comes at the cost of old capability, a phenomenon known as catastrophic forgetting. This area studies how a system can keep learning throughout its deployment while retaining what it already knew, and how it should decide when adaptation is actually warranted rather than reacting to noise.
Why It Matters
Systems deployed for long periods will encounter conditions their training never covered. A system that can only be updated by full retraining is fragile and expensive to maintain compared with one that can adapt online.
Core Concepts
- Catastrophic forgetting
- Elastic weight consolidation and related regularization methods
- Experience replay
- Task boundary detection
- Plasticity versus stability tradeoff
Relevant Disciplines
- Neuroscience of memory consolidation
- Developmental psychology
- Cognitive science
Potential Mimicry Sources
- Sleep dependent memory consolidation
- Lifelong skill retention in animals
- Human skill retention after long disuse
Projects
No projects yet.
Active Questions
No projects yet, so there are no active derived questions to report.
Key Findings Across Projects
Pending. No projects in this area have produced findings yet.
Unresolved Questions
Pending.
Connections to Other Areas
Pending. Connections will be identified as projects in this area develop.