Machine Unlearning for Masked Diffusion Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, Georu, Jeong, Seungwon, Kim, Hoki, Park, Jinseong, Lee, Woojin
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918509209976832
author Lee, Georu
Jeong, Seungwon
Kim, Hoki
Park, Jinseong
Lee, Woojin
author_facet Lee, Georu
Jeong, Seungwon
Kim, Hoki
Park, Jinseong
Lee, Woojin
contents Recent masked diffusion language models (MDLMs), such as LLaDA and Dream, have achieved performance comparable to autoregressive large language models. Unlike autoregressive models, which generate text sequentially, MDLMs generate text by iteratively denoising masked positions in parallel. During fine-tuning, MDLMs learn to recover responses from masked response states conditioned on a prompt, thereby shifting their predictions from a prompt-masked unconditional distribution toward a prompt-conditional distribution. Despite this distinct generative and fine-tuning mechanism, machine unlearning for MDLMs remains largely unexplored. In this paper, we propose Masked Diffusion Unlearning (MDU), the first unlearning framework for MDLMs, by revisiting the process of learning specific knowledge in terms of diffusion. Specifically, MDU minimizes a forward KL divergence from the prompt-conditional prediction to a prompt-masked unconditional anchor at every masked response position, with a temperature scaling parameter to control the privacy-utility trade-off. Our empirical results on standard benchmarks and MDLM backbones show that MDU achieves high unlearning performance compared to existing LLM unlearning methods. Code is available at https://github.com/leegeoru/MDU.
format Preprint
id arxiv_https___arxiv_org_abs_2605_18253
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Machine Unlearning for Masked Diffusion Language Models
Lee, Georu
Jeong, Seungwon
Kim, Hoki
Park, Jinseong
Lee, Woojin
Computation and Language
Artificial Intelligence
I.2.7; I.2.6
Recent masked diffusion language models (MDLMs), such as LLaDA and Dream, have achieved performance comparable to autoregressive large language models. Unlike autoregressive models, which generate text sequentially, MDLMs generate text by iteratively denoising masked positions in parallel. During fine-tuning, MDLMs learn to recover responses from masked response states conditioned on a prompt, thereby shifting their predictions from a prompt-masked unconditional distribution toward a prompt-conditional distribution. Despite this distinct generative and fine-tuning mechanism, machine unlearning for MDLMs remains largely unexplored. In this paper, we propose Masked Diffusion Unlearning (MDU), the first unlearning framework for MDLMs, by revisiting the process of learning specific knowledge in terms of diffusion. Specifically, MDU minimizes a forward KL divergence from the prompt-conditional prediction to a prompt-masked unconditional anchor at every masked response position, with a temperature scaling parameter to control the privacy-utility trade-off. Our empirical results on standard benchmarks and MDLM backbones show that MDU achieves high unlearning performance compared to existing LLM unlearning methods. Code is available at https://github.com/leegeoru/MDU.
title Machine Unlearning for Masked Diffusion Language Models
topic Computation and Language
Artificial Intelligence
I.2.7; I.2.6
url https://arxiv.org/abs/2605.18253