While large-scale text-to-image diffusion models have demonstrated impressive
image-generation capabilities, there are significant concerns about their
potential misuse for generating unsafe content, violating copyright, and
perpetuating societal biases. Recently, the text-to-image generation community
has begun addressing these concerns by editing or unlearning undesired concepts
from pre-trained models. However, these methods often involve data-intensive
and inefficient fine-tuning or utilize various forms of token remapping,
rendering them susceptible to adversarial jailbreaks. In this paper, we present
a simple and effective training-free approach, ConceptPrune, wherein we first
identify critical regions within pre-trained models responsible for generating
undesirable concepts, thereby facilitating straightforward concept unlearning
via weight pruning. Experiments across a range of concepts including artistic
styles, nudity, object erasure, and gender debiasing demonstrate that target
concepts can be efficiently erased by pruning a tiny fraction, approximately
0.12% of total weights, enabling multi-concept erasure and robustness against
various white-box and black-box adversarial attacks.

我们提出了一种简单而有效的无需训练的方法，ConceptPrune，通过首先确定在预训练模型中负责生成不良概念的关键区域，从而以权重修剪的方式便捷地实现概念去学习。实验证明，我们的方法能够高效擦除多个目标概念，仅修剪总权重的约 0.12%，并对多种白盒和黑盒对抗攻击具有鲁棒性。

ConceptPrune: 通过技能神经元修剪在扩散模型中进行概念编辑

ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron  Pruning

Large Generative AI (GAI) models have the unparalleled ability to generate
text, images, audio, and other forms of media that are increasingly
indistinguishable from human-generated content. As these models often train on
publicly available data, including copyrighted materials, art and other
creative works, they inadvertently risk violating copyright and
misappropriation of intellectual property (IP). Due to the rapid development of
generative AI technology and pressing ethical considerations from stakeholders,
protective mechanisms and techniques are emerging at a high pace but lack
systematisation.
In this paper, we study the concerns regarding the intellectual property
rights of training data and specifically focus on the properties of generative
models that enable misuse leading to potential IP violations. Then we propose a
taxonomy that leads to a systematic review of technical solutions for
safeguarding the data from intellectual property violations in GAI.

大型生成型人工智能（GAI）模型可以生成逐渐无法区分是否人工生成的文本、图片、声音和其他形式的媒体。本文研究了训练数据的知识产权问题，重点关注生成模型的特性，探讨可能导致潜在知识产权侵犯的滥用行为，并提出了一个分类体系，对 GAI 中保护数据免受知识产权侵犯的技术解决方案进行系统评述。

生成型人工智能中数据的知识产权保护方法评述

U Can't Gen This? A Survey of Intellectual Property Protection Methods  for Data in Generative AI

With the rapid development of Large Language Models (LLMs), we have witnessed
intense competition among the major LLM products like ChatGPT, LLaMa, and
Gemini. However, various issues (e.g. privacy leakage and copyright violation)
of the training corpus still remain underexplored. For example, the Times sued
OpenAI and Microsoft for infringing on its copyrights by using millions of its
articles for training. From the perspective of LLM practitioners, handling such
unintended privacy violations can be challenging. Previous work addressed the
``unlearning" problem of LLMs using gradient information, while they mostly
introduced significant overheads like data preprocessing or lacked robustness.
In this paper, contrasting with the methods based on first-order information,
we revisit the unlearning problem via the perspective of second-order
information (Hessian). Our unlearning algorithms, which are inspired by classic
Newton update, are not only data-agnostic/model-agnostic but also proven to be
robust in terms of utility preservation or privacy guarantee. Through a
comprehensive evaluation with four NLP datasets as well as a case study on
real-world datasets, our methods consistently show superiority over the
first-order methods.

通过利用 Hessian 的二阶信息，我们提出的消除大型语言模型中的隐私侵犯问题的方法在效用保护和隐私保证方面经过全面评估和真实数据集的案例研究后， consistently show superiority over the first-order methods。