In real-world applications, a machine learning model is required to handle an open-set recognition (OSR), where unknown classes appear during the inference, in addition to a domain shift, where the distribution of data differs between the training and inference phases. Domain generalization (DG) aims to handle the domain shift situation where the target domain of the inference phase is inaccessible during model training. Open domain generalization (ODG) takes into account both DG and OSR. Domain-Augmented Meta-Learning (DAML) is a method targeting ODG but has a complicated learning process. On the other hand, although various DG methods have been proposed, they have not been evaluated in ODG situations. This work comprehensively evaluates existing DG methods in ODG and shows that two simple DG methods, CORrelation ALignment (CORAL) and Maximum Mean Discrepancy (MMD), are competitive with DAML in several cases. In addition, we propose simple extensions of CORAL and MMD by introducing the techniques used in DAML, such as ensemble learning and Dirichlet mixup data augmentation. The experimental evaluation demonstrates that the extended CORAL and MMD can perform comparably to DAML with lower computational costs. This suggests that the simple DG methods and their simple extensions are strong baselines for ODG. The code used in the experiments is available at https://github.com/shiralab/OpenDG-Eval.
翻译:在实际应用中,机器学习模型需要处理开放式识别 (OSR) 的情况,即除了培训过程中未知类别的出现外,还必须处理培训和识别阶段数据分布不同的情况。领域泛化 (DG) 旨在处理培训过程中不可访问的推理阶段的目标域的情况。开放域泛化 (ODG) 考虑了 DG 和 OSR。领域增强元学习 (DAML) 是面向 ODG 的方法,但具有复杂的学习过程。另一方面,尽管已经提出了各种各样的 DG 方法,但它们在 ODG 情况下尚未得到评估。本研究全面评估现有的 DG 方法在 ODG 中的表现,并显示两种简单的 DG 方法,即 CORrelation ALignment (CORAL) 和 Maximum Mean Discrepancy (MMD),在若干情况下与 DAML 竞争。此外,我们通过引入 DAML 中使用的技术,如集合学习和 Dirichlet mixup 数据增广,提出了 CORAL 和 MMD 的简单扩展。实验评估表明,扩展的 CORAL 和 MMD 可以以更低的计算成本与 DAML 相当。这表明,简单的 DG 方法及其简单的扩展是 ODG 的强基线。实验中使用的代码可在 https://github.com/shiralab/OpenDG-Eval 上获得。