Reproducibility Challenge: Making AI Forget You: Data Deletion in Machine Learning

Xiaohui Wang, Zijin Nie, Zijun Yu

Abstract:

This report examines the reproducibility of the paper Making AI Forget You: Data Deletion In Machine Learning.
The original paper initiated a framework studying what to do when specific data is no longer accessible for deploying models. The paper also proposed two efficient deletion algorithms for k-means clustering model called Q-k-means and DC-k-means. We studied the replicability of the two algorithms, examining if they could achieve similar performance as they stated in the report. By re-implementing Q-k-means and DC-k-means, we reproduced the experiments using MNIST, Gaussian and Covtype datasets. We also performed the experiments of the effect of parameter width and lambda to the efficiency of algorithms. The experiments we conducted yield similar results as shown in the original paper.

Track: Replicability

NeurIPS Paper Id: https://openreview.net/forum?id=H1xovVBg8B