Catastrophic forgetting

Catastrophic forgetting is when a neural network trained on a new task loses much of what it learned on an earlier one.

In plain words

A network stores what it knows in its weights. Training on a new task changes those same weights, which can overwrite the settings that made the old task work. One fix slows down changes to the weights that matter most for earlier tasks.

A quick example

Imagine a model that reads handwritten digits well. It is then trained only on a second task. Tested on digits again, its accuracy may have dropped sharply, because the new training reshaped weights the digit skill relied on.

Where you'll meet it

• Continual learning, where a system learns tasks one after another.
• Research on learning a series of games, such as Atari titles, in sequence.

Related

Sources

1. PaperAn Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks, Goodfellow et al. (arXiv) · read 29 Sept 2026
2. PaperOvercoming catastrophic forgetting in neural networks, Kirkpatrick et al. (arXiv) · read 29 Sept 2026

Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.