The paper "In Search of Lost Domain Generalization" by Ishaan Gulrajani and David Lopez-Paz from Facebook AI Research addresses the challenge of domain generalization, which aims to predict well on distributions different from those seen during training. The authors highlight the importance of model selection criteria in domain generalization tasks and argue that algorithms without a well-defined model selection strategy are incomplete. They implement DOMAINBED, a testbed for domain generalization, which includes seven multi-domain datasets, nine baseline algorithms, and three model selection criteria. Extensive experiments using DOMAINBED reveal that empirical risk minimization (ERM) achieves state-of-the-art performance when equipped with modern neural network architectures and data augmentation techniques. The authors also release DOMAINBED to facilitate reproducible and rigorous research in domain generalization, encouraging contributions from the research community. The paper concludes with a discussion on future research directions, emphasizing the need for more realistic datasets and the importance of model selection methods.The paper "In Search of Lost Domain Generalization" by Ishaan Gulrajani and David Lopez-Paz from Facebook AI Research addresses the challenge of domain generalization, which aims to predict well on distributions different from those seen during training. The authors highlight the importance of model selection criteria in domain generalization tasks and argue that algorithms without a well-defined model selection strategy are incomplete. They implement DOMAINBED, a testbed for domain generalization, which includes seven multi-domain datasets, nine baseline algorithms, and three model selection criteria. Extensive experiments using DOMAINBED reveal that empirical risk minimization (ERM) achieves state-of-the-art performance when equipped with modern neural network architectures and data augmentation techniques. The authors also release DOMAINBED to facilitate reproducible and rigorous research in domain generalization, encouraging contributions from the research community. The paper concludes with a discussion on future research directions, emphasizing the need for more realistic datasets and the importance of model selection methods.