arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2304.01973cs.LGcs.CV
收藏

ERM++:一种改进的域泛化基线

ERM++: An Improved Baseline for Domain Generalization

Boston University(波士顿大学);MIT Lincoln Laboratory(麻省理工学院林肯实验室)

Piotr Teterwak, Kuniaki Saito, Theodoros Tsiligkaridis, Kate Saenko, Bryan A. Plummer

AI总结:

本文提出 ERM++,通过调优训练数据利用、初始化和正则化等被忽视因素,在 DomainBed 上以简单基线显著超越 ERM 及现有 DG 方法。

摘要:

域泛化(DG)旨在开发能够泛化到新的、未见过的数据分布的分类器,当收集特定于新域的数据不切实际时,这是一项关键能力。一个常见的 DG 基线会最小化源域上的经验风险。近期研究表明,这种被称为经验风险最小化(Empirical Risk Minimization,ERM)的方法在经过适当调优时,可以胜过大多数更复杂的 DG 方法。然而,这些研究主要集中在一组狭窄的超参数上,忽略了其他能够增强鲁棒性并防止过拟合和灾难性遗忘的因素,而这些特性对于出色的 DG 性能至关重要。在我们对训练数据利用(即持续时间和设置验证集划分)、初始化以及额外正则化项的研究中,我们发现调优这些此前被忽视的因素,可在不增加太多复杂性的情况下显著提升模型在多种数据集上的泛化能力。我们将这种改进但简单的基线称为 ERM++。尽管易于实现,在包含 5 个数据集的标准基准上,使用 ResNet-50 时,ERM++ 相比以往 ERM 基线将 DG 性能提高了 5% 以上,使用 ViT-B/16 时则提高了 15% 以上。在两种架构下,它还在 DomainBed 数据集上优于所有最先进方法。重要的是,ERM++ 很容易集成到 DomainBed 等现有框架中,使其成为研究人员和从业者实用且强大的工具。总体而言,ERM++ 通过提供一个更强大、更可靠且保持简单易用的基线,挑战了对更复杂 DG 方法的需求。代码可在 https://github.com/piotr-teterwak/erm_plusplus 获取。

英文摘要:

Domain Generalization (DG) aims to develop classifiers that can generalize to new, unseen data distributions, a critical capability when collecting new domain-specific data is impractical. A common DG baseline minimizes the empirical risk on the source domains. Recent studies have shown that this approach, known as Empirical Risk Minimization (ERM), can outperform most more complex DG methods when properly tuned. However, these studies have primarily focused on a narrow set of hyperparameters, neglecting other factors that can enhance robustness and prevent overfitting and catastrophic forgetting, properties which are critical for strong DG performance. In our investigation of training data utilization (i.e., duration and setting validation splits), initialization, and additional regularizers, we find that tuning these previously overlooked factors significantly improves model generalization across diverse datasets without adding much complexity. We call this improved, yet simple baseline ERM++. Despite its ease of implementation, ERM++ improves DG performance by over 5\% compared to prior ERM baselines on a standard benchmark of 5 datasets with a ResNet-50 and over 15\% with a ViT-B/16. It also outperforms all state-of-the-art methods on DomainBed datasets with both architectures. Importantly, ERM++ is easy to integrate into existing frameworks like DomainBed, making it a practical and powerful tool for researchers and practitioners. Overall, ERM++ challenges the need for more complex DG methods by providing a stronger, more reliable baseline that maintains simplicity and ease of use. Code is available at \url{https://github.com/piotr-teterwak/erm_plusplus}

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。

↑