Centrala begrepp
Proposing a novel domain generalization framework, WDRDG, to address the challenge of limited labeled samples in source domains by leveraging Wasserstein distributional robust optimization.
Sammanfattning
The content discusses the challenges of domain generalization and proposes a novel framework, WDRDG, to address these challenges. It introduces the concept of Wasserstein uncertainty sets and optimal transport for adaptive inference. The framework is evaluated on three datasets: VLCS, PACS, and Rotated MNIST, showing superior performance in handling unseen domain shifts with limited training data.
JOURNAL OF L AT EX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021
Introduction to the concept of domain generalization and its challenges.
Proposal of a novel domain generalization framework called WDRDG.
Explanation of Wasserstein uncertainty sets and optimal transport for adaptive inference.
Evaluation of the framework on VLCS, PACS, and Rotated MNIST datasets.
Data Extraction:
"Experiments on the Rotated MNIST, PACS and the VLCS datasets demonstrate that our method could effectively balance the robustness and discriminability in challenging generalization scenarios."
Statistik
Experiments on the Rotated MNIST, PACS and the VLCS datasets demonstrate that our method could effectively balance the robustness and discriminability in challenging generalization scenarios.
Citat
"We propose a domain generalization framework that solves the Wasserstein distributionally robust optimization problem to learn a robust model over multiple source domains."
"Our main contributions include proposing a domain generalization framework that solves the Wasserstein distributionally robust optimization problem."