How do unsupervised learning systems handle catastrophic forgetting?

They use replay buffers, regularization, and parameter isolation to preserve learned representations.

Catastrophic forgetting occurs when models overwrite previously learned representations when trained on new data. Unsupervised systems mitigate this using experience replay (storing embeddings), elastic weight consolidation (penalizing important weight changes), and modular architectures like mixture-of-experts. These strategies ensure stability while still adapting to new distributions.