TASK	DATASET	MODEL	METRIC NAME	METRIC VALUE	GLOBAL RANK	REMOVE
Domain Adaptation	Office-31	ELS	Average Accuracy	90.4	# 9
Domain Adaptation	Office-Home	ELS	Accuracy	84.6	# 5

Badge	Markdown
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/free-lunch-for-domain-adversarial-training-1/domain-adaptation-on-office-home)](https://paperswithcode.com/sota/domain-adaptation-on-office-home?p=free-lunch-for-domain-adversarial-training-1)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/free-lunch-for-domain-adversarial-training-1/domain-adaptation-on-office-31)](https://paperswithcode.com/sota/domain-adaptation-on-office-31?p=free-lunch-for-domain-adversarial-training-1)`

Free Lunch for Domain Adversarial Training: Environment Label Smoothing

The Eleventh International Conference on Learning Representations (ICLR 2023) 2023 · Yifan Zhang, Xue Wang, Jian Liang, Zhang Zhang, Liang Wang, Rong Jin, Tieniu Tan ·

A fundamental challenge for machine learning models is how to generalize learned models for out-of-distribution (OOD) data. Among various approaches, exploiting invariant features by Domain Adversarial Training (DAT) received widespread attention. Despite its success, we observe training instability from DAT, mostly due to over-confident domain discriminator and environment label noise. To address this issue, we proposed Environment Label Smoothing (ELS), which encourages the discriminator to output soft probability, which thus reduces the confidence of the discriminator and alleviates the impact of noisy environment labels. We demonstrate, both experimentally and theoretically, that ELS can improve training stability, local convergence, and robustness to noisy environment labels. By incorporating ELS with DAT methods, we are able to yield state-of-art results on a wide range of domain generalization/adaptation tasks, particularly when the environment labels are highly noisy.

PDF Abstract The Eleventh 2023 PDF