Erasure or Erosion? Evaluating Compositional Degradation in Unlearned Text-To-Image Diffusion Models

• URL: http://arxiv.org/abs/2604.04575v1
• Date: Mon, 06 Apr 2026 10:16:42 GMT
• Title: Erasure or Erosion? Evaluating Compositional Degradation in Unlearned Text-To-Image Diffusion Models
• Abstract summary: Post-hoc unlearning has emerged as a practical mechanism for removing undesirable concepts from large text-to-image diffusion models.<n>We conduct a systematic empirical study of concept unlearning through the lens of compositional text-to-image generation.<n>Our results reveal a consistent trade-off between unlearning effectiveness and compositional integrity.
• Score: 7.251531287880316
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Post-hoc unlearning has emerged as a practical mechanism for removing undesirable concepts from large text-to-image diffusion models. However, prior work primarily evaluates unlearning through erasure success; its impact on broader generative capabilities remains poorly understood. In this work, we conduct a systematic empirical study of concept unlearning through the lens of compositional text-to-image generation. Focusing on nudity removal in Stable Diffusion 1.4, we evaluate a diverse set of state-of-the-art unlearning methods using T2I-CompBench++ and GenEval, alongside established unlearning benchmarks. Our results reveal a consistent trade-off between unlearning effectiveness and compositional integrity: methods that achieve strong erasure frequently incur substantial degradation in attribute binding, spatial reasoning, and counting. Conversely, approaches that preserve compositional structure often fail to provide robust erasure. These findings highlight limitations of current evaluation practices and underscore the need for unlearning objectives that explicitly account for semantic preservation beyond targeted suppression.

Related papers

• Continual Unlearning for Text-to-Image Diffusion Models: A Regularization Perspective [35.50502807526103]
  We present the first systematic study of continual unlearning in text-to-image diffusion models.<n>We show that popular unlearning methods suffer from rapid utility collapse after only a few requests.<n>We propose a gradient-projection method that constrains parameter drift to their subspace.
  arXiv  Detail & Related papers  (2025-11-11T08:33:16Z)
• Semantic Surgery: Zero-Shot Concept Erasure in Diffusion Models [27.672305802461377]
  We introduce a novel training-free, zero-shot framework for concept erasure that operates directly on text embeddings before the diffusion process.<n>We achieve superior completeness and robustness while preserving locality and image quality.<n>This robustness also allows our framework to function as a built-in threat detection system, offering a practical solution for safer text-to-image generation.
  arXiv  Detail & Related papers  (2025-10-26T22:04:17Z)
• Sculpting Memory: Multi-Concept Forgetting in Diffusion Models via Dynamic Mask and Concept-Aware Optimization [20.783312940122297]
  Text-to-image (T2I) diffusion models have achieved remarkable success in generating high-quality images from textual prompts.<n>However, their ability to store vast amounts of knowledge raises concerns in scenarios where selective forgetting is necessary.<n>We propose textbfDynamic Mask coupled with Concept-Aware Loss, a novel unlearning framework designed for multi-concept forgetting.
  arXiv  Detail & Related papers  (2025-04-12T01:38:58Z)
• Fine-Grained Erasure in Text-to-Image Diffusion-based Foundation Models [56.35484513848296]
  FADE (Fine grained Attenuation for Diffusion Erasure) is an adjacency-aware unlearning algorithm for text-to-image generative models.<n>It removes target concepts with minimal impact on correlated concepts, achieving a 12% improvement in retention performance over state-of-the-art methods.
  arXiv  Detail & Related papers  (2025-03-25T15:49:48Z)
• TRCE: Towards Reliable Malicious Concept Erasure in Text-to-Image Diffusion Models [53.937498564603054]
  Recent advances in text-to-image diffusion models enable photorealistic image generation, but they also risk producing malicious content, such as NSFW images.<n>To mitigate risk, concept erasure methods are studied to facilitate the model to unlearn specific concepts.<n>We propose TRCE, using a two-stage concept erasure strategy to achieve an effective trade-off between reliable erasure and knowledge preservation.
  arXiv  Detail & Related papers  (2025-03-10T14:37:53Z)
• Holistic Unlearning Benchmark: A Multi-Faceted Evaluation for Text-to-Image Diffusion Model Unlearning [11.566380065176526]
  Concept unlearning is a promising solution to unethical or harmful use of text-to-image diffusion models.<n>Our benchmark covers 33 target concepts, including 16,000 prompts per concept, spanning four categories: Celebrity, Style, Intellectual Property, and NSFW.<n>Our investigation reveals that no single method excels across all evaluation criteria.
  arXiv  Detail & Related papers  (2024-10-08T03:30:39Z)
• Six-CD: Benchmarking Concept Removals for Benign Text-to-image Diffusion Models [58.74606272936636]
  Text-to-image (T2I) diffusion models have shown exceptional capabilities in generating images that closely correspond to textual prompts.<n>The models could be exploited for malicious purposes, such as generating images with violence or nudity, or creating unauthorized portraits of public figures in inappropriate contexts.<n> concept removal methods have been proposed to modify diffusion models to prevent the generation of malicious and unwanted concepts.
  arXiv  Detail & Related papers  (2024-06-21T03:58:44Z)
• Unlearning Concepts in Diffusion Model via Concept Domain Correction and Concept Preserving Gradient [20.698305103879232]
  We propose a novel concept domain correction framework named textbfDoCo (textbfDomaintextbfCorrection)<n>By aligning the output domains of sensitive and anchor concepts through adversarial training, our approach ensures comprehensive unlearning of target concepts.<n>We also introduce a concept-preserving gradient surgery technique that mitigates conflicting gradient components, thereby preserving the model's utility while unlearning specific concepts.
  arXiv  Detail & Related papers  (2024-05-24T07:47:36Z)
• Introspective Deep Metric Learning [91.47907685364036]
  We propose an introspective deep metric learning framework for uncertainty-aware comparisons of images. The proposed IDML framework improves the performance of deep metric learning through uncertainty modeling.
  arXiv  Detail & Related papers  (2023-09-11T16:21:13Z)
• PAIF: Perception-Aware Infrared-Visible Image Fusion for Attack-Tolerant Semantic Segmentation [50.556961575275345]
  We propose a perception-aware fusion framework to promote segmentation robustness in adversarial scenes. We show that our scheme substantially enhances the robustness, with gains of 15.3% mIOU, compared with advanced competitors.
  arXiv  Detail & Related papers  (2023-08-08T01:55:44Z)
• Incremental False Negative Detection for Contrastive Learning [95.68120675114878]
  We introduce a novel incremental false negative detection for self-supervised contrastive learning. During contrastive learning, we discuss two strategies to explicitly remove the detected false negatives. Our proposed method outperforms other self-supervised contrastive learning frameworks on multiple benchmarks within a limited compute.
  arXiv  Detail & Related papers  (2021-06-07T15:29:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.