Joint Distribution Optimal Transportation for Domain Adaptation

Abstract

This paper deals with the unsupervised domain adaptation problem, where one wants to estimate a prediction function $f$ in a given target domain without any labeled sample by exploiting the knowledge available from a source domain where labels are known. Our work makes the following assumption: there exists a non-linear transformation between the joint feature/label space distributions of the two domain ${\mathcal{P}}_s$ and ${\mathcal{P}}_t$. We propose a solution of this problem with optimal transport, that allows to recover an estimated target ${\mathcal{P}}_t^f=(X,f(X))$ by optimizing simultaneously the optimal coupling and $f$. We show that our method corresponds to the minimization of a bound on the target error, and provide an efficient algorithmic solution, for which convergence is proved. The versatility of our approach, both in terms of class of hypothesis or loss functions is demonstrated with real world classification and regression problems, for which we reach or surpass state-of-the-art results.

arXiv commentAccepted for publication at NIPS 2017

References in corpus (3)

• DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition
• Domain Adaptation: Learning Bounds and Algorithms
• A Transportation $L^p$ Distance for Signal Analysis

Cited by in corpus (20)

• Large-Scale Optimal Transport and Mapping Estimation
• 2-Wasserstein Approximation via Restricted Convex Potentials with Application to Improved Training for GANs
• Causal Generative Domain Adaptation Networks
• A Survey on Optimal Transport for Machine Learning: Theory and Applications
• Towards Unsupervised Domain Adaptation via Domain-Transformer
• Prototypical Partial Optimal Transport for Universal Domain Adaptation
• Deep Active Learning: Unified and Principled Method for Query and Training
• A Primer on Domain Adaptation
• Multi-Source Domain Adaptation through Dataset Dictionary Learning in Wasserstein Space
• Stylized Face Sketch Extraction via Generative Prior with Limited Data
• Adversarial Bi-Regressor Network for Domain Adaptive Regression
• Unsupervised Domain Adaptation Based on Source-guided Discrepancy
• Adaptation Across Extreme Variations using Unlabeled Domain Bridges
• On the Wasserstein median of probability measures
• Understanding the Limits of Unsupervised Domain Adaptation via Data Poisoning
• Review and Examination of Input Feature Preparation Methods and Machine Learning Models for Turbulence Modeling
• On the Robustness of Domain Adaption to Adversarial Attacks
• Principled Representation Learning for Entity Alignment
• Adversarial training for predictive tasks: theoretical analysis and limitations in the deterministic case
• Dynamic Flows on Curved Space Generated by Labeled Data