Preserving Linear Separability in Continual Learning by Backward Feature Projection

Open in new window