arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.09022cs.LG
收藏

一种改进近OOD检测的Mahalanobis距离修正方法

A Simple Fix to Mahalanobis Distance for Improving Near-OOD Detection

Google Research(谷歌研究院);Stanford University(斯坦福大学);Harvard University(哈佛大学);Google Health(谷歌健康)

Jie Ren, Stanislav Fort, Jeremiah Liu, Abhijit Guha Roy, Shreyas Padhy, Balaji Lakshminarayanan

摘要:

马氏距离(MD)是检测神经网络中分布外(OOD)输入的一种简单而流行的后处理方法。我们分析了它的故障模式近OOD检测,并提出了一个简单的修复称为相对马氏距离(RMD),提高了性能,更强大的超参数选择。在广泛选择的具有挑战性的视觉,语言和生物学OOD基准(CIFAR-100与CIFAR-10,CLINC OOD意图检测,基因组学OOD)上,我们表明RMD有意义地提高了MD性能(在基因组学OOD上高达15%的AUROC)。

英文摘要:

Mahalanobis distance (MD) is a simple and popular post-processing method for detecting out-of-distribution (OOD) inputs in neural networks. We analyze its failure modes for near-OOD detection and propose a simple fix called relative Mahalanobis distance (RMD) which improves performance and is more robust to hyperparameter choice. On a wide selection of challenging vision, language, and biology OOD benchmarks (CIFAR-100 vs CIFAR-10, CLINC OOD intent detection, Genomics OOD), we show that RMD meaningfully improves upon MD performance (by up to 15% AUROC on genomics OOD).

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。

↑