Title: PECKER: A Precisely Efficient Critical Knowledge Erasure Recipe For Machine Unlearning in Diffusion Models

Original: arXiv:2604.05634v2 Announce Type: replace Abstract: Machine unlearning (MU) has become a critical technique for GenAI models' safe and compliant operation. While existing MU methods are effective, most impose prohibitive training time and computational overhead. Our analysis suggests the root cause lies in poorly directed gradient updates, which reduce training efficiency and destabilize convergence. To mitigate these issues, we propose PECKER, an efficient MU approach that matches or outperforms prevailing methods. Within a distillation framework, PECKER introduces a saliency mask to prioritize updates to parameters that contribute most to forgetting the targeted data, thereby reducing unnecessary gradient computation and shortening overall training time without sacrificing unlearning efficacy. Our method generates samples that unlearn related class or concept more quickly, while closely aligning with the true image distribution on CIFAR-10 and STL-10 datasets, achieving shorter training times for both class forgetting and concept forgetting.

Rewrite: Machine unlearning (MU) is now an essential component for ensuring that generative AI models operate securely and in compliance with regulations. Although current unlearning techniques are proven to work, they often demand excessive computational resources and extend training durations significantly. Our investigation indicates that this inefficiency stems from gradient updates that lack precise direction, leading to slower training speeds and unstable convergence patterns. To address these challenges, we introduce PECKER, a highly efficient unlearning strategy that performs on par with or better than existing state-of-the-art solutions. Operating within a distillation-based framework, PECKER employs a saliency mask to focus parameter updates on those most responsible for erasing the targeted information. This targeted approach eliminates redundant gradient calculations, thereby accelerating the overall training process while maintaining strong unlearning performance. On the CIFAR-10 and STL-10 datasets, our technique demonstrates faster unlearning of specific classes or concepts while preserving alignment with the underlying image distribution, ultimately reducing the time required for both class and concept erasure.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC