Discriminative Radial Domain Adaptation

Abstract

Domain adaptation methods reduce domain shift typically by learning domain-invariant features. Most existing methods are built on distribution matching, e.g., adversarial domain adaptation, which tends to corrupt feature discriminability. In this paper, we propose Discriminative Radial Domain Adaptation (DRDR) which bridges source and target domains via a shared radial structure. It's motivated by the observation that as the model is trained to be progressively discriminative, features of different categories expand outwards in different directions, forming a radial structure. We show that transferring such an inherently discriminative structure would enable to enhance feature transferability and discriminability simultaneously. Specifically, we represent each domain with a global anchor and each category a local anchor to form a radial structure and reduce domain shift via structure matching. It consists of two parts, namely isometric transformation to align the structure globally and local refinement to match each category. To enhance the discriminability of the structure, we further encourage samples to cluster close to the corresponding local anchors based on optimal-transport assignment. Extensively experimenting on multiple benchmarks, our method is shown to consistently outperforms state-of-the-art approaches on varied tasks, including the typical unsupervised domain adaptation, multi-source domain adaptation, domainagnostic learning, and domain generalization.

Unsupervised Subdomain Adaptation Robust to Noisy Pseudo Labels Using Symmetric Loss

“…Pseudo label, as a semi-supervised learning method, has recently become an effective and competitive method for obtaining target domain category information in UDA tasks [15,[26][27][28]. Zhang et al proposed ProDA [15] method, in which soft pseudo labels are online corrected based on the relative feature distance from the prototype.…”

“…Similar methods include Enhancing Transferability and Discriminability Simultaneously (ETDS) [35]， Backprop Induced Feature Weighting for Adversarial Domain Adaptation (BIWAA) [26] ， Discriminative Radial Domain Adaptation (DRDA) [28]. Environment Label Smoothing (ELS) [36] method, which was recently published in ICLR 2023, attempts to enhance its robustness to environmental noise in domain adaptation models.…”

“…Wang et al presented Self-adaptive Re-weighted Adversarial Domain Adaptation [27], which utilizes source samples and pseudo labeled target samples to achieve better inter-class separation and intra-class compactness of features. Huang et al proposed DRDA [28] method, which utilizes pseudo labels to obtain local anchors of the target domain.…”