Correlation Alignment for Unsupervised Domain Adaptation

By Baochen Sun, Jiashi Feng, Kate Saenko. Domain Adaptation in Computer Vision Applications, 2016

512 citations recorded by the source corpus.

In this chapter, we present CORrelation ALignment (CORAL), a simple yet effective method for unsupervised domain adaptation. CORAL minimizes domain shift by aligning the second-order statistics of source and target distributions, without requiring any target labels. In contrast to subspace manifold methods, it aligns the original feature distributions of the source and target domains, rather than the bases of lower-dimensional subspaces. It is also much simpler than other distribution matching methods. CORAL performs remarkably well in extensive evaluations on standard benchmark datasets. We first describe a solution that applies a linear transformation to source features to align them with target features before classifier training. For linear classifiers, we propose to equivalently apply CORAL to the classifier weights, leading to added efficiency when the number of classifiers is small but the number and dimensionality of target examples are very high. The resulting CORAL Linear Discriminant Analysis (CORAL-LDA) outperforms LDA by a large margin on standard domain adaptation benchmarks. Finally, we extend CORAL to learn a nonlinear transformation that aligns correlations of layer activations in deep neural networks (DNNs). The resulting Deep CORAL approach works seamlessly with DNNs and achieves state-of-the-art performance on standard benchmark datasets. Our code is available at: https://github.com/VisionLearningGroup/CORAL.

Subfields: Artificial Intelligence

View Paper: https://www.semanticscholar.org/paper/d4a196e67e9e47a9797670b7e4ae59f2330baca0

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

Citations in the corpus

Improving robustness against common corruptions by covariate shift adaptation (2020, background): “[47] and follow up work evaluated the performance of adapting BN parameters in unsupervised [5, 6] and supervised DA settings [4].”
Attending to Discriminative Certainty for Domain Adaptation (2019, background): “Other work in[16] and [42] have used Cycle consistency [55] loss and deep coral loss [41, 40] for ensuring the closeness of source and target domain respectively.”
Beyond Sharing Weights for Deep Domain Adaptation (2016, methodology): “A similar idea was followed in [44] to minimize the difference in second order feature statistics using the CORAL loss [45] and to relate higher order feature statistics in [46].”
Tent: Fully Test-Time Adaptation by Entropy Minimization (2021, background): “These losses optimize feature alignment (Gretton et al., 2009; Sun et al., 2017), adversarial invariance (Ganin & Lempitsky, 2015; Tzeng et al., 2017), or shared proxy tasks (Sun et al., 2019a).”
SqueezeSegV2: Improved Model Structure and Unsupervised Domain Adaptation for Road-Object Segmentation from a LiDAR Point Cloud (2018, methodology): “Another category of relevant work employs a discrepancy loss [15], [16], [17], [6], which explicitly measures the discrepancy between the source and target domains on corresponding activation layers of the two network streams.”
Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift (2020, methodology): “, 2016], match means and covariances across feature vectors [Sun et al., 2017], or match moments of the distributions directly [Peng et al.”
Incremental Adversarial Domain Adaptation for Continually Changing Environments (2017, background): “[18] align second order statistics of layer activations for source and target domains.”
TTT++: When Does Self-Supervised Test-Time Training Fail or Thrive? (2021, methodology): “More speciﬁcally, we use the square distance of the ﬁrst and second moments between two feature distributions, inspired by DDC [30] and Coral [22], to approximate the domain discrepancy.”