Towards Efficient Target-Level Machine Unlearning Based on Essential Graph
Heng Xu, Tianqing Zhu, Lefeng Zhang, Wanlei Zhou, Wei Zhao
TL;DR
This work addresses target-level unlearning, where the goal is to erase the influence of a specific target within multi-target instances without harming remaining targets. It introduces an essential graph to capture inter-layer parameter relationships and a Grad-CAM-inspired parameter selection process, followed by a balanced graph construction to jointly consider all targets. By pruning only the most critical, target-linked parameters, the method achieves efficient unlearning with minimal degradation to remaining performance, and it demonstrates robustness against model-inversion and membership inference attacks. The approach significantly reduces computational and storage overhead compared with retraining from scratch, and it shows versatility across classification, segmentation, and detection tasks, with strong practical privacy implications for real-world deployments.
Abstract
Machine unlearning is an emerging technology that has come to attract widespread attention. A number of factors, including regulations and laws, privacy, and usability concerns, have resulted in this need to allow a trained model to forget some of its training data. Existing studies of machine unlearning mainly focus on unlearning requests that forget a cluster of instances or all instances from one class. While these approaches are effective in removing instances, they do not scale to scenarios where partial targets within an instance need to be forgotten. For example, one would like to only unlearn a person from all instances that simultaneously contain the person and other targets. Directly migrating instance-level unlearning to target-level unlearning will reduce the performance of the model after the unlearning process, or fail to erase information completely. To address these concerns, we have proposed a more effective and efficient unlearning scheme that focuses on removing partial targets from the model, which we name "target unlearning". Specifically, we first construct an essential graph data structure to describe the relationships between all important parameters that are selected based on the model explanation method. After that, we simultaneously filter parameters that are also important for the remaining targets and use the pruning-based unlearning method, which is a simple but effective solution to remove information about the target that needs to be forgotten. Experiments with different training models on various datasets demonstrate the effectiveness of the proposed approach.