Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Spartalis, Christoforos N., Semertzidis, Theodoros, Daras, Petros, Gavves, Efstratios
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2508.20773
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918132028801024
author Spartalis, Christoforos N.
Semertzidis, Theodoros
Daras, Petros
Gavves, Efstratios
author_facet Spartalis, Christoforos N.
Semertzidis, Theodoros
Daras, Petros
Gavves, Efstratios
contents We introduce SAFEMax, a novel method for Machine Unlearning in diffusion models. Grounded in information-theoretic principles, SAFEMax maximizes the entropy in generated images, causing the model to generate Gaussian noise when conditioned on impermissible classes by ultimately halting its denoising process. Also, our method controls the balance between forgetting and retention by selectively focusing on the early diffusion steps, where class-specific information is prominent. Our results demonstrate the effectiveness of SAFEMax and highlight its substantial efficiency gains over state-of-the-art methods.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20773
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Unleashing Uncertainty: Efficient Machine Unlearning for Generative AI
Spartalis, Christoforos N.
Semertzidis, Theodoros
Daras, Petros
Gavves, Efstratios
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
We introduce SAFEMax, a novel method for Machine Unlearning in diffusion models. Grounded in information-theoretic principles, SAFEMax maximizes the entropy in generated images, causing the model to generate Gaussian noise when conditioned on impermissible classes by ultimately halting its denoising process. Also, our method controls the balance between forgetting and retention by selectively focusing on the early diffusion steps, where class-specific information is prominent. Our results demonstrate the effectiveness of SAFEMax and highlight its substantial efficiency gains over state-of-the-art methods.
title Unleashing Uncertainty: Efficient Machine Unlearning for Generative AI
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.20773