Deep structure alignment network for scalable unsupervised domain adaptation

doi:10.1016/j.knosys.2026.115641

Deep structure alignment network for scalable unsupervised domain adaptation

2026-02-27

0

PRE

AI

DOI：10.1016/j.knosys.2026.115641

Original

Original request for help

Save

Abstract

En 中文

• A unified Deep Structure Alignment Network (DSAN) is designed for cross-domain adaptation, effectively improving deep model transferability by aligning global structural information. • Scalable global alignment achieved via a sparse anchor mechanism, resolving the conflict between mini-batch training efficiency and structural coherence. • Computationally efficient optimization using a differentiable spectral loss that avoids expensive eigendecomposition, making the method applicable to large-scale datasets.

Keywords:

Deep Structure Alignment Network

Unsupervised Domain Adaptation

Global Structural Alignment

Sparse Anchor Mechanism

Differentiable Spectral Loss

Deep structure alignment network for scalable unsupervised domain adaptation

Abstract

Journal

Organization