Score-Based Generative Classifiers
Roland S. Zimmermann、Lukas Schott1、Yang Song2、Benjamin A. Dunn3、David A. Klindt3 University of Tübingen, 2 Stanford University, 3 Norwegian University of Science and Technology.
Abstract
The tremendous success of generative models in recent years raises the question whether they can also be used to perform classification. Generative models have been used as adversarially robust classifiers on simple datasets such as MNIST, but this robustness has not been observed on more complex datasets like CIFAR-10. Additionally, on natural image datasets, previous results have suggested a trade-off between the likelihood of the data and classification accuracy. In this work, we investigate score-based generative models as classifiers for natural images. We show that these models not only obtain competitive likelihood values but simultaneously achieve state-of-the-art classification accuracy for generative classifiers on CIFAR-10. Nevertheless, we find that these models are only slightly, if at all, more robust than discriminative baseline models on out-of-distribution tasks based on common image corruptions. Similarly and contrary to prior results, we find that score-based are prone to worst-case distribution shifts in the form of adversarial perturbations. Our work highlights that score-based generative models are closing the gap in classification accuracy compared to standar
中文速览
生成式模型（generative model）能不能同时做好图像分类，一直是个悬而未决的问题——过去的研究普遍认为，在复杂自然图像上，模型的数据似然（likelihood）越高，分类精度就越低，二者难以兼得。本文提出用基于分数的生成模型（score-based generative classifier, SBGC）来做分类，具体做法是在扩散模型的分数函数中加入类别条件，对每张图片计算各类别的条件似然，取最大值作为预测标签。实验表明，该方法在CIFAR-10上同时实现了生成式分类器中最高的分类精度和极具竞争力的似然值，打破了此前"精度与似然必须权衡"的结论，并且对像素空间插值图像能正确地给出较低似然，表现出更合理的分布外感知能力。然而，在面对常见图像噪声污染（CIFAR-10-C）时，该模型的提升仅限于噪声类扰动，对其他类型腐蚀并无优势；更关键的是，在对抗样本攻击下，模型对所有被攻击图像全部分类错误，与在MNIST上表现出鲁棒性的早期生成式分类器形成鲜明反差。这说明，尽管SBGC在干净数据上的准确率已接近判别式模型，但生成式分类路线在自然图像上的对抗鲁棒性问题仍远未解决，值得深入探索。
原文 arXiv:2110.00473;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2110.00473v2