Learning to Unlearn: Machine Unlearning via Learning the Unlearning Behaviors
The paper proposes learning a lightweight unlearning function instead of hand-designing one for each removal task.
L2UL treats machine unlearning behavior as something to learn from a distribution, aiming to approximate retraining on `D \ D_f` without the same compute burden. The authors say existing unlearning functions can become a bottleneck on large datasets even when the model is not especially large. Their experiments report retraining-comparable accuracy with better efficiency, including scalability checks on larger ResNet models. ArXiv · AI/CL/LG's note
L2UL treats machine unlearning behavior as something to learn from a distribution, aiming to approximate retraining on `D \ D_f` without the same compute burden. The authors say existing unlearning functions can become a bottleneck on large datasets even when the model is not especially large. Their experiments report retraining-comparable accuracy with better efficiency, including scalability checks on larger ResNet models. ArXiv · AI/CL/LG's note
score 4