Joint cross-domain classification and subspace learning for unsupervised\n adaptation

2014/11/17 by Basura Fernando, Fernando, Basura, Tatiana Tommasi +3
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #Cancer-related molecular mechanisms research #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Machine Learning (cs.LG) #Respiratory viral infections research

paper · pdf · doi:10.48550/arxiv.1411.4491

openalex publication_date 2014/11/17 · openalex created_date 2022/10/01 · openalex updated_date 2026/07/28

Abstract

Domain adaptation aims at adapting the knowledge acquired on a source domain\nto a new different but related target domain. Several approaches have\nbeenproposed for classification tasks in the unsupervised scenario, where no\nlabeled target data are available. Most of the attention has been dedicated to\nsearching a new domain-invariant representation, leaving the definition of the\nprediction function to a second stage. Here we propose to learn both jointly.\nSpecifically we learn the source subspace that best matches the target subspace\nwhile at the same time minimizing a regularized misclassification loss. We\nprovide an alternating optimization technique based on stochastic sub-gradient\ndescent to solve the learning problem and we demonstrate its performance on\nseveral domain adaptation tasks.\n

Citations

• Domain Adaptation: Learning Bounds and Algorithms
• Visual Domain Adaptation: A survey of recent advances

Related

• Unsupervised Domain Adaptation: A Multi-task Learning-based Method
• Cross-domain error minimization for unsupervised domain adaptation
• Cross Domain Adaptation by Learning Partially Shared Classifiers and Weighting Source Data Points in the Shared Subspaces
• Unsupervised Domain Adaptation with Progressive Domain Augmentation
• Low-Rank Subspace Override for Unsupervised Domain Adaptation