不要偏爱多数:扩散模型的少数群体引导
Don't Play Favorites: Minority Guidance for Diffusion Models
AI总结:
本文提出minority guidance采样技术,通过度量样本独特性并引导扩散模型生成过程聚焦低密度区域,从而高效生成高质量少数样本,并在医学成像等真实场景中验证了其有效性。
摘要:
我们探索使用扩散模型生成少数样本的问题。少数样本是位于数据流形低密度区域上的实例。生成足够数量的此类少数实例非常重要,因为它们通常包含数据的一些独特属性。然而,扩散模型的传统生成过程由于其高似然而大多产生多数样本(位于流形高密度区域上的样本),这使得它们对于少数生成任务而言效率低下且耗时。在这项工作中,我们提出了一个新的框架,可以使扩散模型的生成过程聚焦于少数样本。我们首先强调,Tweedie去噪公式对多数样本产生有利的结果。这一观察促使我们引入一个度量,用于描述给定样本的独特性。为了解决扩散模型对多数样本的固有偏好,我们进一步开发了minority guidance,这是一种采样技术,可以将生成过程引导至具有期望似然水平的区域。在基准真实数据集上的实验表明,我们的minority guidance能够显著提高生成高质量少数样本的能力,优于现有的生成采样器。我们展示了我们的框架的性能优势即使在要求苛刻的现实世界场景(如医学成像)中也持续存在,进一步强调了我们工作的实际意义。代码可在 https://github.com/soobin-um/minority-guidance 获取。
英文摘要:
We explore the problem of generating minority samples using diffusion models. The minority samples are instances that lie on low-density regions of a data manifold. Generating a sufficient number of such minority instances is important, since they often contain some unique attributes of the data. However, the conventional generation process of the diffusion models mostly yields majority samples (that lie on high-density regions of the manifold) due to their high likelihoods, making themselves ineffective and time-consuming for the minority generating task. In this work, we present a novel framework that can make the generation process of the diffusion models focus on the minority samples. We first highlight that Tweedie's denoising formula yields favorable results for majority samples. The observation motivates us to introduce a metric that describes the uniqueness of a given sample. To address the inherent preference of the diffusion models w.r.t. the majority samples, we further develop minority guidance, a sampling technique that can guide the generation process toward regions with desired likelihood levels. Experiments on benchmark real datasets demonstrate that our minority guidance can greatly improve the capability of generating high-quality minority samples over existing generative samplers. We showcase that the performance benefit of our framework persists even in demanding real-world scenarios such as medical imaging, further underscoring the practical significance of our work. Code is available at https://github.com/soobin-um/minority-guidance.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。