DeepJDOT: Deep Joint distribution optimal transport for unsupervised domain adaptation

By Bharath Bhushan Damodaran, Benjamin Kellenberger, Rémi Flamary, Devis Tuia, Nicolas Courty. 2018

541 citations recorded by the source corpus.

In computer vision, one is often confronted with problems of domain shifts, which occur when one applies a classifier trained on a source dataset to target data sharing similar characteristics (e.g. same classes), but also different latent data structures (e.g. different acquisition conditions). In such a situation, the model will perform poorly on the new data, since the classifier is specialized to recognize visual cues specific to the source domain. In this work we explore a solution, named DeepJDOT, to tackle this problem: through a measure of discrepancy on joint deep representations/labels based on optimal transport, we not only learn new data representations aligned between the source and target domain, but also simultaneously preserve the discriminative information used by the classifier. We applied DeepJDOT to a series of visual recognition tasks, where it compares favorably against state-of-the-art deep domain adaptation methods.

Subfields: Artificial Intelligence, Cancer Research, Computer Vision and Pattern Recognition

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

Citations in the corpus

Unsupervised Model Adaptation for Continual Semantic Segmentation (2020, methodology): “Wasserstein Distance (WD) has been used successfully for domain alignment in the UDA literature (Bhushan Damodaran et al. 2018b; Rostami et al. 2019; Xu et al. 2020; Li et al. 2020; Rostami and Galstyan 2020).”
A Review of Domain Adaptation without Target Labels (2019, background): “distance is important to generative adversarial networks, and there have been a number of works combining optimal transport with deep learning [137], [138], [139].”
Universal Domain Adaptation (2019, background): “Note that the entropy is normalized by its maximum value (log |Cs|) so that it is restricted into [0, 1] and comparable to the domain similarity measure d̂′.”
SAR Image Classification Using Few-Shot Cross-Domain Transfer Learning (2019, background): “Optimal transport [33] is a suitable metric to deal with the above issues and due to be an effective metric, it has been used extensively in deep learning literature re-Algorithm Update entire parameters: end for cently [5, 4, 15, 28].”
A Survey of Unsupervised Deep Domain Adaptation (2018, background): “While domain adaptation methods commonly align feature distributions, a different line of research aligns the joint or conditional distribution of the feature and label spaces instead [44, 47, 50, 148, 151, 154, 233, 267].”
Sliced Wasserstein Discrepancy for Unsupervised Domain Adaptation (2019, methodology): “It is interesting to see that the task-specific discrepancyaware methods such as MCD [57], DeepJDOT [12], and the proposed SWD are the current leading approaches for the tasks being addressed here.”
Deep Transfer Learning for Few-Shot SAR Image Classification (2019, background): “For this reason, it has been found to be an effective metric and has 292 been used extensively in deep learning literature recently [15,22,28,29].”
Sequential Unsupervised Domain Adaptation through Prototypical Distributions (2020, methodology): “Damodaran et al. [21] use the WD for domain alignment in a UDA setting which leads to considerable performance improvement compared to methods [10, 19] that rely on matching lower order statistics.”
Multi-Adversarial Discriminative Deep Domain Generalization for Face Presentation Attack Detection (2019, methodology): “The straightforward way to solve this problem is to exploit the domain adaptation technique [27, 12, 25, 20, 34, 7, 24, 29, 6, 30, 33, 3] to align the feature distribution between training and testing data so that the trained model with source data can be adapted on the target data.”
Enhanced Transport Distance for Unsupervised Domain Adaptation (2020, background): “Recently, Bhushan et al. [3] employ deep neural networks to explore the optimal transport plan and further learn the domain-similar features.”
Improving and generalizing flow-based generative models with minibatch optimal transport (2023, unknown): “Although minibatch OT incurs an error relative to the exact OT solution, it has been successfully used in many applications like domain adaptation or generative modeling (Damodaran et al., 2018; Genevay et al., 2018).”
Learning Transferable Knowledge Through Embedding Spaces (2019, methodology): “Note that we cannot even approximate eC(θ∗) in the UDA framework as there is no labeled data in the target domain.”
Partially Does It: Towards Scene-Level FG-SBIR with Partial Input (2022, background): “Later works extended optimal transport to the deep learning landscape for applications such as document classification [33], fewshot learning [80], domain adaptation [17], self-supervised learning [45], neural machine translation [75], and comprehending scene using 3D point cloud from monocular data”
Deep Unsupervised Domain Adaptation: A Review of Recent Advances and Perspectives (2022, methodology): “As a deep learning framework, DeepJDOT is further proposed to achieve an end-to-end UDA (Damodaran et al., 2018).”
Domain Adaptive Ensemble Learning (2020, background): “In particular, domaininvariant features can be learned by directly minimizing distribution divergence measures, such as MMD [33,35,36] and optimal transport [7,2].”
Learning a Domain-Invariant Embedding for Unsupervised Domain Adaptation Using Class-Conditioned Distribution Alignment (2019, unknown): “In contrast, more recent techniques that use a shared deep encoder, employ the Wasserstein metric [42] to address UDA [4], [6].”
BAPA-Net: Boundary Adaptation and Prototype Alignment for Cross-domain Semantic Segmentation (2021, unknown): “Many previous works try to minimize the Maximum Mean Discrepancy [20, 36, 34], KL-divergence [47], optimal transport distance [8, 9], etc.”
Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation (2020, background): “…paradigms, moment matching and adversarial alignment, as aforementioned, there are several sample-based DA methods that estimate the importance weights of source samples (Jiang & Zhai, 2007; Huang et al., 2007) or sample-to-sample similarity via optimal transport (Bhushan Damodaran et al., 2018).”