Catastrophic Forgetting
A problem where a machine learning model loses previously learned knowledge after being trained on new data, reducing performance on earlier tasks.
What is Catastrophic Forgetting?
Catastrophic forgetting commonly occurs in neural networks trained sequentially on multiple tasks. As the model updates its parameters to learn a new task, those changes can overwrite information that was important for earlier tasks. As a result, performance on previously learned tasks may decline significantly, even while performance on the new task improves.
Why is Catastrophic Forgetting Important?
AI systems may need to continuously learn from new information without losing existing capabilities. Catastrophic forgetting makes this difficult and can reduce model reliability over time. Addressing it is particularly important for continual learning systems that must adapt to changing data, environments, or requirements.
Common use cases
Catastrophic forgetting is commonly studied in continual learning, robotics, autonomous systems, personalized AI, incremental model training, and neural networks that learn multiple tasks over time.