Continual learning is the problem of teaching a model something new without wiping out what it already knew. Variational quantum classifiers suffer this badly: train them on a sequence of tasks whose distribution shifts and earlier skills degrade, the failure usually called catastrophic forgetting.

A standard classical remedy is elastic weight consolidation, which works out which parameters mattered for previous tasks and makes them harder to move. Deciding what counts as important is the crux, and the classical version measures it with classical Fisher information.

The proposal here is quantum elastic weight consolidation, which uses quantum Fisher information instead. The argument is that a quantum model's parameters live in a geometry the classical measure does not describe properly, so importance should be judged by a measure native to that geometry. Same protective instinct as the classical method, with a ruler that fits the space being measured.