arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2303.09746cs.LGcs.CV
收藏

通过类别条件印象的再现来检测分布外的示例

Detecting Out-of-distribution Examples via Class-conditional Impressions Reappearing

Huazhong University of Science and Technology(华中科技大学);Ping An Technology (Shenzhen) Co., Ltd.(平安科技(深圳)有限公司)

Jinggang Chen, Xiaoyang Qu, Junjie Li, Jianzong Wang, Jiguang Wan, Jing Xiao

摘要:

分布外(OOD)检测旨在增强标准深度神经网络,以区分异常输入和原始训练数据。以前的进展已经介绍了各种方法,其中分布训练数据甚至几个OOD示例是先决条件。然而,由于隐私和安全性,辅助数据在现实世界中往往是不切实际的。在本文中,我们提出了一种无需对自然数据进行训练的无数据方法,称为类条件印象再现(C2IR),它利用固定模型的图像印象来恢复类条件特征统计。在此基础上,我们引入积分概率估计来估计逐层的类条件偏差,并通过基于重要性的度量(MGI)来获得层权重。实验验证了该方法的有效性,并表明C2IR优于其他事后方法,达到了与完全访问(ID和OOD)检测方法相当的性能,特别是在远OOD数据集(SVHN)。

英文摘要:

Out-of-distribution (OOD) detection aims at enhancing standard deep neural networks to distinguish anomalous inputs from original training data. Previous progress has introduced various approaches where the in-distribution training data and even several OOD examples are prerequisites. However, due to privacy and security, auxiliary data tends to be impractical in a real-world scenario. In this paper, we propose a data-free method without training on natural data, called Class-Conditional Impressions Reappearing (C2IR), which utilizes image impressions from the fixed model to recover class-conditional feature statistics. Based on that, we introduce Integral Probability Metrics to estimate layer-wise class-conditional deviations and obtain layer weights by Measuring Gradient-based Importance (MGI). The experiments verify the effectiveness of our method and indicate that C2IR outperforms other post-hoc methods and reaches comparable performance to the full access (ID and OOD) detection method, especially in the far-OOD dataset (SVHN).

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。

↑