Abstract

In this work, we propose a domain generalization (DG) approach to learn on several labeled source domains and transfer knowledge to a target domain that is inaccessible in training. Considering the inherent conditional and label shifts, we would expect the alignment of p (x| y) and p (y). However, the widely used domain invariant feature learning (IFL) methods relies on aligning the marginal concept shift wrt p (x), which rests on an unrealistic assumption that p (y) is invariant across domains. We thereby propose a novel variational Bayesian inference framework to enforce the conditional distribution alignment wrt p (x| y) via the prior distribution matching in a latent space, which also takes the marginal label shift wrt p (y) into consideration with the posterior alignment. Extensive experiments on various benchmarks demonstrate that our framework is robust to the label shift and the crossdomain accuracy is significantly improved, thereby achieving superior performance over the conventional IFL counterparts.

Date

2021

Authors

Xiaofeng Liu, Bo Hu, Linghao Jin, Xu Han, Fangxu Xing1 Jinsong Ouyang1 Jun Lu, Georges El Fakhri1 Jonghye Woo

Journal

Proc. Int. Joint Conf. Artif. Intell

Pages

881-887