Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning