The widely used domain invariant feature learning (IFL) methods relies on aligning the marginal concept shift w.r.t. $p(x|y)$ and the marginal label shift . 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 . We thereby propose a novelvariational Bayesian inference framework to enforce the conditionaldistribution alignment . The framework is robust to the label shift and thecross-domain accuracy is significantly improved, thereby achieving superiorperformance over the conventional IFL counterparts. Extensive experiments on variousbenchmarks demonstrate that our framework is

Author(s) : Xiaofeng Liu, Bo Hu, Linghao Jin, Xu Han, Fangxu Xing, Jinsong Ouyang, Jun Lu, Georges EL Fakhri, Jonghye Woo

Links : PDF - Abstract

Code :
Coursera

Keywords : domain - framework - label - shift - bayesian -

Leave a Reply

Your email address will not be published. Required fields are marked *