Optimal Transport for Domain Adaptation

By Nicolas Courty, Rémi Flamary, Devis Tuia, Alain Rakotomamonjy. 2014

1,364 citations recorded by the source corpus.

Domain adaptation is one of the most challenging tasks of modern data analytics. If the adaptation is done correctly, models built on a specific data representation become more robust when confronted to data depicting the same classes, but described by another observation system. Among the many strategies proposed, finding domain-invariant representations has shown excellent properties, in particular since it allows to train a unique classifier effective in all domains. In this paper, we propose a regularized unsupervised optimal transportation model to perform the alignment of the representations in the source and target domains. We learn a transportation plan matching both PDFs, which constrains labeled samples of the same class in the source domain to remain close during transport. This way, we exploit at the same time the labeled samples in the source and the distributions observed in both domains. Experiments on toy and challenging real visual adaptation examples show the interest of the method, that consistently outperforms state of the art approaches. In addition, numerical experiments show that our approach leads to better performances on domain invariant deep learning features and can be easily adapted to the semi-supervised case where few labeled samples are available in the target domain.

Subfields: Artificial Intelligence, Cancer Research, 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 Cancer Research

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Affine Invariant Analysis of Frank-Wolfe on Strongly Convex Sets (2020, background): “When the optimum is in the relative interior of any compact set C , FW converges linearly when f is strongly convex [Gu´elat et al. 1986; Lacoste-Julien et al. 2013].”
Co-Transport for Class-Incremental Learning (2021, neutral): “model fusion [37], domain adaptation [7], object detection [10], model reuse [56, 57], and generative models [2, 4].”
Unsupervised Model Adaptation for Continual Semantic Segmentation (2020, background): “The Wasserstein distance (WD) (Courty et al. 2016; Bhushan Damodaran et al. 2018a) is an example of such measures which captures higher-order statistics.”
Generative Continual Concept Learning (2019, methodology): “For this reason, we select the Wasserstein Distance (WD) metric (Bonnotte 2013) which satisfies this requirement and has recently been used extensively in deep learning applications to measure the distance between two probability distributions (Courty et al. 2017).”
Domain Adaptation for Visual Applications: A Comprehensive Survey (2017, methodology): “Similarly, the Optimal Transport for Domain Adaptation (OTDA) [46], considers a local transportation plan for each source.”
Theoretical Analysis of Domain Adaptation with Optimal Transport (2016, methodology): “Thus, we conclude that the regularized transport formulations such as the one of [6,7] can be seen as algorithmic solutions for controlling the trade-off between the terms of the bound.”
DeepJDOT: Deep Joint distribution optimal transport for unsupervised domain adaptation (2018, methodology): “Optimal transport [22–24] has been used in domain adaptation to learn the transformation between domains [4, 25, 26], with associated theoretical guarantees [27].”
Comprehensive survey on hierarchical clustering algorithms and the recent developments (2022, methodology): “…a new agglomerative hierarchical clustering algorithm called HC-OT, hierarchical clustering is combined with optimal transport (OT) (Rabin et al. 2014; Courty et al. 2017) for the first time, and the OT based distance is also proposed to be similarity measure used in HC-OT, this new distance…”
Multi-Domain Adversarial Learning (2019, methodology): “Accordingly, a large set of DA methods proceed by reducing this distance in the original input space X , e.g. via importance sampling (Bickel et al., 2007) or by modifying the source representation using optimal transport (Courty et al., 2017; Damodaran et al., 2018).”
Learning Transferable Knowledge Through Embedding Spaces (2019, methodology): “[30] used a regularized version of the optimal transport for domain adaptation to circumvent computational complexity.”
Distant Supervised Centroid Shift: A Simple and Efficient Approach to Visual Domain Adaptation (2019, background): “over the last decade have been devoted to transfer learning [64, 8, 34, 36] and multi-task learning [14, 30, 63, 70] that leverage the latent relationship with previous datasets to learn a new model for an emerging dataset.”
Projection-Free Optimization on Uniformly Convex Sets (2020, background): “…al., 2017; Miech et al., 2018), quadrature rules in RKHS (Bach et al., 2012; Lacoste-Julien et al., 2015; Futami et al., 2019), optimal transport (Courty et al., 2016; Vayer et al., 2018; Paty and Cuturi, 2019; Luise et al., 2019), and many others. x t +1 = (1 − γ t ) x t + γ t v t (cid:46)…”
OTCE: A Transferability Metric for Cross-Domain Cross-Task Representations (2021, methodology): “The entropic regularized OT has been used for domain adaptation to compute the optimal mapping of input from the source domain to the target domain [10].”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, neutral): “Many previous works try to minimize the Maximum Mean Discrepancy [20, 36, 34], KL-divergence [47], optimal transport distance [8, 9], etc.”
Deep Transfer Learning for Few-Shot SAR Image Classification (2019, methodology): “[22] proved the following theorem in standard domain adaptation which provides an upper-bound on 324 the target domain risk given the source domain risk and the joint combined risk.”
Fairness Violations and Mitigation under Covariate Shift (2019, methodology): “• OTDA is an optimal transport-based method for unsupervised domain adaptation [14].”
On Localized Discrepancy for Domain Adaptation (2020, methodology): “Courty et al. (2017); Redko et al. (2017); Courty et al. (2017) consider using Wasserstein distance as surrogate of the parameterized discrepancy.”