Dataset Distillation for Domain Generalization
Keywords: Dataset Distillation, Domain Generalization, Style Transfer, Self-Supervised Learning
TL;DR: Propose a novel task, Dataset Distillation For Domain Generalizaiton, which aims to provide distilled dataset that can be train a model robust to unseen domain generalization.
Abstract: Dataset Distillation (DD) has been applied to various downstream tasks and recently scaled to ImageNet-1k, highlighting its potential for practical applications. However, in real-world scenarios, robustness to unseen domains is essential, and the robustness of models trained on synthetic datasets remains uncertain. To address this, we propose a novel task, Dataset Distillation for Domain Generalization (DD for DG), and evaluate the unseen domain generalization of models trained on synthetic datasets distilled by state-of-the-art DD methods using the DomainBed benchmark. Additionally, we introduce a new method for this task, which interprets DD through the lens of image style transfer, achieving superior performance in unseen domain generalization compared to baseline approaches.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8996
Loading