Learning to Generate Novel Domains for Domain Generalization

By Kaiyang Zhou, Yongxin Yang, Timothy M. Hospedales, Tao A. Xiang. European Conference on Computer Vision, 2020

557 citations recorded by the source corpus.

This paper focuses on domain generalization (DG), the task of learning from multiple source domains a model that generalizes well to unseen domains. A main challenge for DG is that the available source domains often exhibit limited diversity, hampering the model's ability to learn to generalize. We therefore employ a data generator to synthesize data from pseudo-novel domains to augment the source domains. This explicitly increases the diversity of available training domains and leads to a more generalizable model. To train the generator, we model the distribution divergence between source and synthesized pseudo-novel domains using optimal transport, and maximize the divergence. To ensure that semantics are preserved in the synthesized data, we further impose cycle-consistency and classification losses on the generator. Our method, L2A-OT (Learning to Augment by Optimal Transport) outperforms current state-of-the-art DG methods on four benchmark datasets.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/ac780229c64422398b0aa212489be2c965d224ce

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Yu Qiao
2. Tao A. Xiang
3. Kaiyang Zhou
4. Ziwei Liu
5. Chen Change Loy
6. Yongxin Yang
7. P. Heng
8. Qi-Yan Dou

In Computer Vision and Pattern Recognition

1. Tao A. Xiang
2. Yu Qiao
3. Chen Change Loy
4. Luc Van Gool
5. Yi Yang
6. Xiaogang Wang
7. Dacheng Tao
8. Qi Tian

Citations in the corpus

• Federated Learning from Pre-Trained Models: A Contrastive Learning Approach (2022, methodology): “[21] Kaiyang Zhou, Yongxin Yang, Timothy Hospedales, and Tao Xiang.”
• FedDG: Federated Domain Generalization on Medical Image Segmentation via Episodic Learning in Continuous Frequency Space (2021, methodology): “There are other methods tackling DG by manipulating deep neural network architectures [19, 23, 35], leveraging self-supervision signals [3, 54], designing training heuristics [17, 25], or conducting data augmentations [48, 53, 60, 61], which are free from requirement of data centralization.”
• Domain Generalization: A Survey (2021, background): “cation [23], [35], but also in other domains such as speech recognition [40], natural language processing [34], medical imaging [41], [42], and reinforcement learning [23].”
• Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization (2021, background): “…discrepancy minimization (Li et al., 2020), adversarial training (Rah-man et al., 2020; Zhao et al., 2020c), invariance representation learning (Zhou et al., 2020; Piratla et al., 2020), etc. Recently, there is significant interest on conducting DG with one source domain only, for which…”
• Learning to Diversify for Single Domain Generalization (2021, unknown): “, alignment-based [2, 11, 33] and augmentation-based [5, 17, 44, 50, 47].”
• Generalizing to Unseen Domains: A Survey on Domain Generalization (2021, background): “[43] generated novel distributions under semantic consistency and then maximized the difference between source and the novel distributions.”
• Domain Adaptive Ensemble Learning (2020, methodology): “To overcome the domain shift problem and improve the generalization ability of DNNs, two related problems have been studied extensively: unsupervised domain adaptation (UDA) [2, 3, 4, 5, 6, 7, 8, 9] and domain generalization (DG) [10, 11, 12, 13, 14, 15, 16].”
• Domain Generalization with MixStyle (2021, methodology): “Whereas L2A-OT works at the pixel level: it trains an image generator by maximizing the domain difference (measured by optimal transport) between the original and the generated images, which introduces much heavier computational overhead than MixStyle in terms of GPU memory and training time.”
• RobustNet: Improving Domain Generalization in Urban-Scene Segmentation via Instance Selective Whitening (2021, background): “Various approaches such as meta-learning [27, 2, 28, 33], adversarial training [29, 32, 46], autoencoder [15, 29], metric learning [10, 39], data augmentation [64, 16, 67] have been proposed to learn domain-agnostic feature representations.”
• Deep Stable Learning for Out-Of-Distribution Generalization (2021, background): “Several works propose to enlarge the available data space with augmentation of source domains [6, 49, 59, 44, 64, 63].”
• A Fourier-based Framework for Domain Generalization (2021, methodology): “Compared with the SOTA, our FACT clearly beats the methods based on adversarial data augmentation or meta learning, including the latest MASF [4], DDAIG [50] and L2A-OT [51], yet FACT enjoys an efficient training process without any additional adversarial or episodic training steps.”