Training

Catastrophic Forgetting

Catastrophic forgetting occurs when fine-tuning a model on new data causes performance degradation on old tasks. The model's weights shift to optimize new data at the expense of previous knowledge. This is a major challenge in continual learning and fine-tuning. Mitigations include: regularization (keep weights close to original), replay (include old data), and architectural approaches. Understanding forgetting is crucial for multi-task learning. High-quality fine-tuning attempts to minimize forgetting.