Deep CORAL: Correlation Alignment for Deep Domain Adaptation

By Baochen Sun, Kate Saenko. 2016

3,996 citations recorded by the source corpus.

Deep neural networks are able to learn powerful representations from large quantities of labeled input data, however they cannot always generalize well across changes in input distributions. Domain adaptation algorithms have been proposed to compensate for the degradation in performance due to domain shift. In this paper, we address the case when the target domain is unlabeled, requiring unsupervised adaptation. CORAL is a "frustratingly easy" unsupervised domain adaptation method that aligns the second-order statistics of the source and target distributions with a linear transformation. Here, we extend CORAL to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks (Deep CORAL). Experiments on standard benchmark datasets show state-of-the-art performance.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised Learning

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

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Break-It-Fix-It: Unsupervised Learning for Program Repair (2021, background): “Within domain adaptation, our task is also related to the setting where unlabeled data in the test domain is available (Sun & Saenko, 2016; Hoffman et al., 2018; Sun et al., 2020) (in our case, real bad code).”
Adversarial Discriminative Domain Adaptation (2017, methodology): “This is generally achieved by optimizing the representation to minimize some measure of domain shift such as maximum mean discrepancy [5, 6] or correlation distances [7, 8].”
Few-Shot Adversarial Domain Adaptation (2017, methodology): “In [54], which is a deep UDA approach, unlabeled target data is used to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks.”
SWAD: Domain Generalization by Seeking Flat Minima (2021, methodology): “Previous approaches explicitly reduced domain gaps in the latent space [8– Table 1: Comparisons with SOTA.”
Unsupervised Model Adaptation for Continual Semantic Segmentation (2020, methodology): “2018a) demonstrate that using WD leads to performance improvement over methods that rely on matching lower-order statistics (Long et al. 2015; Sun and Saenko 2016).”
CyCADA: Cycle-Consistent Adversarial Domain Adaptation (2017, background): “Alignment typically involves minimizing some measure of distance between the source and target feature distributions, such as maximum mean discrepancy (Long & Wang, 2015), correlation distance (Sun & Saenko, 2016), or adversarial discriminator accuracy (Ganin & Lempitsky, 2015; Tzeng et al., 2017).”
In Search of Lost Domain Generalization (2020, background): “– CORAL [Sun and Saenko, 2016] matches the mean and covariance of feature distributions.”
Class-Incremental Domain Adaptation (2020, background): “The most popular UDA approach [11,25,30,46,49,52] is to learn a predictor h(x) = g ◦ f(x) having a domain-agnostic feature extractor f that is common to both the domains, and a classifier g which can be learned using source supervision.”
Unified Deep Supervised Domain Adaptation and Generalization (2017, methodology): “In [56], which is a UDA approach, unlabeled target data are used to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks.”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “n negative constant during the backpropagation-based training to ensures that the feature distributions over the two domains are made indistinguishable. (Image courtesy Y. Ganin [95]). The Deep CORAL [218] extends the shallow CORAL [216] method described in Section ?? to deep architectures33. The ma”
WILDS: A Benchmark of in-the-Wild Distribution Shifts (2020, positive): “Numerous methods have been proposed for domain generalization, including domain-invariant learning (Ganin et al., 2016; Sun and Saenko, 2016), invariant risk minimization (IRM) (Arjovsky et al., 2019), and meta-learning-based methods (Li et al., 2017; Dou et al., 2019).”
Contrastive Adaptation Network for Unsupervised Domain Adaptation (2019, background): “A common practice for domain adaptation is to minimize the discrepancy be-tween domains in order to obtain domain-invariant features [9, 4, 21, 18, 20, 31].”
Generalizing to Unseen Domains via Adversarial Data Augmentation (2018, background): “There is a substantial body of work on domain adaptation [15, 3, 32, 9, 39, 36, 26, 40], which aims to better generalize to a priori fixed target domains whose labels are unknown at training time.”
Central Moment Discrepancy (CMD) for Domain-Invariant Representation Learning (2017, methodology): “We further compare our method to LSSA (Aljundi et al., 2015), CORAL (Sun et al., 2016), and AdaBN (Li et al., 2016), based on the ﬁne-tuning of InceptionBN (Ioffe & Szegedy, 2015).”
Revisiting Batch Normalization For Practical Domain Adaptation (2016, methodology): “Recently, Deep CORAL (Sun and Saenko 2016) also extends the method into DNN by incorporating a CORAL loss.”
Domain-Specific Batch Normalization for Unsupervised Domain Adaptation (2019, methodology): “Global alignment techniques include Maximum Mean Discrepancy (MMD) [1, 28], Central Moment Discrepancy (CMD) [29], Wassterstein distance [21], and CORrelation ALignment (CORAL) [22], some of which are reformulated using deep neural networks [12, 23, 28].”
Co-regularized Alignment for Unsupervised Domain Adaptation (2018, methodology): “Domain alignment has almost become a representative approach for domain adaptation, acting as a crucial component in many recently proposed methods [17, 26, 41, 40, 4, 16, 44, 42, 35].”
Multi-Modal Domain Adaptation for Fine-Grained Action Recognition (2019, background): “UDA has been extensively studied for vision tasks including object recognition [14, 15, 32, 33, 51, 55], semantic segmentation [18, 60, 65] and person reidentification [10, 49, 62].”