arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2306.04072cs.LG
收藏

探索基于L2归一化的简单、高质量分布外检测

Exploring Simple, High Quality Out-of-Distribution Detection with L2 Normalization

Simon Fraser University(西蒙菲莎大学);MetaOptima

Jarrod Haas, William Yolland, Bernhard Rabus

AI总结:

本研究提出基于L2归一化的极简分布外检测方法,仅需两行代码、无需额外调优等操作,在多种架构和数据集上验证了其接近最优水平的检测性能与训练效率优势。

摘要:

我们证明,对特征空间进行L2归一化,在部分模型和数据集上可实现出色的分布外(Out-of-Distribution, OoD)检测性能。尽管该方法未达到绝对的最优(state-of-the-art)性能,但其极简性十分突出:仅需新增两行代码,无需专门的损失函数、图像增强、离群点暴露或额外的参数调优。我们还观察到,在部分数据集和架构下,训练效率可能更高。值得注意的是,在CIFAR10数据集上使用ResNet18仅训练60轮(或使用ResNet50训练100轮),其在部分近、远分布外数据集上的AUROC性能与多种最优方法的差距可控制在两个百分点以内。我们为该方法提供了理论与实证支持,并验证了其在5种架构和3个分布内(In-Distribution, ID)数据集上的可行性。

英文摘要:

We demonstrate that L2 normalization over feature space can produce capable performance for Out-of-Distribution (OoD) detection for some models and datasets. Although it does not demonstrate outright state-of-the-art performance, this method is notable for its extreme simplicity: it requires only two addition lines of code, and does not need specialized loss functions, image augmentations, outlier exposure or extra parameter tuning. We also observe that training may be more efficient for some datasets and architectures. Notably, only 60 epochs with ResNet18 on CIFAR10 (or 100 epochs with ResNet50) can produce performance within two percentage points (AUROC) of several state-of-the-art methods for some near and far OoD datasets. We provide theoretical and empirical support for this method, and demonstrate viability across five architectures and three In-Distribution (ID) datasets.

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。

↑