TechPrufer
Get started
Apple
Apple logoApple · Feature trail

Low Influence Point Unlearning

2 posts · Aug 2025 – Aug 2026

What this trail covers

This post introduces a method for unlearning specific data points from trained machine learning models by identifying and leveraging low-influence points. This approach aims to reduce the computational costs associated with unlearning by pre-filtering datasets, leading to significant savings (up to ~50%) on real-world empirical examples.

All Apple trails
  1. When Unlearning Is Free: Leveraging Low Influence Points to Reduce Computational CostsThis post proposes an efficient unlearning framework that identifies subsets of training data with negligible impact on model outputs using influence functions. By reducing the size of datasets before unlearning, it achieves significant computational savings (up to approximately 50 percent) on real-world empirical examples.
  2. When Unlearning Is Free: Leveraging Low Influence Points to Reduce Computational CostsIntroduces a framework for efficient unlearning by identifying and leveraging low-influence points in training data, reducing computational costs by up to 50% through dataset pre-filtering.