Research

Paper

AI LLM March 19, 2026

A Concept is More Than a Word: Diversified Unlearning in Text-to-Image Diffusion Models

Authors

Duc Hao Pham, Van Duy Truong, Duy Khanh Dinh, Tien Cuong Nguyen, Dien Hy Ngo, Tuan Anh Bui

Abstract

Concept unlearning has emerged as a promising direction for reducing the risks of harmful content generation in text-to-image diffusion models by selectively erasing undesirable concepts from a model's parameters. Existing approaches typically rely on keywords to identify the target concept to be unlearned. However, we show that this keyword-based formulation is inherently limited: a visual concept is multi-dimensional, can be expressed in diverse textual forms, and often overlap with related concepts in the latent space, making keyword-only unlearning, which imprecisely indicate the target concept is brittle and prone to over-forgetting. This occurs because a single keyword represents only a narrow point estimate of the concept, failing to cover its full semantic distribution and entangled variations in the latent space. To address this limitation, we propose Diversified Unlearning, a distributional framework that represents a concept through a set of contextually diverse prompts rather than a single keyword. This richer representation enables more precise and robust unlearning. Through extensive experiments across multiple benchmarks and state-of-the-art baselines, we demonstrate that integrating Diversified Unlearning as an add-on component into existing unlearning pipelines consistently achieves stronger erasure, better retention of unrelated concepts, and improved robustness against adversarial recovery attacks.

Metadata

arXiv ID: 2603.18767
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-19
Fetched: 2026-03-20 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.18767v1</id>\n    <title>A Concept is More Than a Word: Diversified Unlearning in Text-to-Image Diffusion Models</title>\n    <updated>2026-03-19T11:20:40Z</updated>\n    <link href='https://arxiv.org/abs/2603.18767v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.18767v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Concept unlearning has emerged as a promising direction for reducing the risks of harmful content generation in text-to-image diffusion models by selectively erasing undesirable concepts from a model's parameters. Existing approaches typically rely on keywords to identify the target concept to be unlearned. However, we show that this keyword-based formulation is inherently limited: a visual concept is multi-dimensional, can be expressed in diverse textual forms, and often overlap with related concepts in the latent space, making keyword-only unlearning, which imprecisely indicate the target concept is brittle and prone to over-forgetting. This occurs because a single keyword represents only a narrow point estimate of the concept, failing to cover its full semantic distribution and entangled variations in the latent space. To address this limitation, we propose Diversified Unlearning, a distributional framework that represents a concept through a set of contextually diverse prompts rather than a single keyword. This richer representation enables more precise and robust unlearning. Through extensive experiments across multiple benchmarks and state-of-the-art baselines, we demonstrate that integrating Diversified Unlearning as an add-on component into existing unlearning pipelines consistently achieves stronger erasure, better retention of unrelated concepts, and improved robustness against adversarial recovery attacks.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-19T11:20:40Z</published>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Duc Hao Pham</name>\n    </author>\n    <author>\n      <name>Van Duy Truong</name>\n    </author>\n    <author>\n      <name>Duy Khanh Dinh</name>\n    </author>\n    <author>\n      <name>Tien Cuong Nguyen</name>\n    </author>\n    <author>\n      <name>Dien Hy Ngo</name>\n    </author>\n    <author>\n      <name>Tuan Anh Bui</name>\n    </author>\n  </entry>"
}