when unlearning is free leveraging low influence points to reduce computational costs
when unlearning is free