Reducing the covariate shift by mirror samples in cross domain alignment

Y Zhao, L Cai - Advances in Neural Information …, 2021 - proceedings.neurips.cc
Advances in Neural Information Processing Systems, 2021proceedings.neurips.cc
Eliminating the covariate shift cross domains is one of the common methods to deal with the
issue of domain shift in visual unsupervised domain adaptation. However, current alignment
methods, especially the prototype based or sample-level based methods neglect the
structural properties of the underlying distribution and even break the condition of covariate
shift. To relieve the limitations and conflicts, we introduce a novel concept named (virtual)
mirror, which represents the equivalent sample in another domain. The equivalent sample …
Abstract
Eliminating the covariate shift cross domains is one of the common methods to deal with the issue of domain shift in visual unsupervised domain adaptation. However, current alignment methods, especially the prototype based or sample-level based methods neglect the structural properties of the underlying distribution and even break the condition of covariate shift. To relieve the limitations and conflicts, we introduce a novel concept named (virtual) mirror, which represents the equivalent sample in another domain. The equivalent sample pairs, named mirror pairs reflect the natural correspondence of the empirical distributions. Then a mirror loss, which aligns the mirror pairs cross domains, is constructed to enhance the alignment of the domains. The proposed method does not distort the internal structure of the underlying distribution. We also provide theoretical proof that the mirror samples and mirror loss have better asymptotic properties in reducing the domain shift. By applying the virtual mirror and mirror loss to the generic unsupervised domain adaptation model, we achieved consistently superior performance on several mainstream benchmarks.
proceedings.neurips.cc
以上显示的是最相近的搜索结果。 查看全部搜索结果