SAFEMax unlearns a class from a diffusion model by training the denoiser to output final-step Gaussian noise for that class, achieving strong forgetting and retention with low compute.
Model sparsity can simplify machine unlearning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unleashing Uncertainty: Efficient Machine Unlearning for Generative AI
SAFEMax unlearns a class from a diffusion model by training the denoiser to output final-step Gaussian noise for that class, achieving strong forgetting and retention with low compute.