A post-training process to remove or reduce the effect of specific data from a model without full retraining.