ViT-MUL: A Baseline Study on Recent Machine Unlearning Methods Applied to Vision Transformers

TL;DR

Comprehensive experiments on ViTs using recent MUL algorithms and datasets are presented and it is anticipated that these experiments, ablation studies, and findings could provide valuable insights and inspire further research in this field.

Abstract

Machine unlearning (MUL) is an arising field in machine learning that seeks to erase the learned information of specific training data points from a trained model. Despite the recent active research in MUL within computer vision, the majority of work has focused on ResNet-based models. Given that Vision Transformers (ViT) have become the predominant model architecture, a detailed study of MUL specifically tailored to ViT is essential. In this paper, we present comprehensive experiments on ViTs using recent MUL algorithms and datasets. We anticipate that our experiments, ablation studies, and findings could provide valuable insights and inspire further research in this field.

Figures (1)

• Figure 1: Visualization results of each unlearning algorithms' performance over the epoch. Each mark in the plot represents the average of 5 seed runs. The first column provides a summary of the trade-off between utility and forgetting for each algorithm.