Adaptive Feature Norm for Unsupervised Subdomain Adaptation

EasyChair Preprint 6704

Abstract

In many real-world problems, obtaining labeled data for a specific machine learning task is expensive. Unsupervised Domain Adaptation (UDA) aims at learning a good predictive model for the target domain using labeled information from the source but only unlabeled samples from the target domain. Most of the previous methods tackle this issue with adversarial methods that contain several loss functions and converge slowly. Recently, subdomain adaptation, which focuses on nuances of the distribution of the relevant subdomains, is getting more and more attention in the UDA field. This paper proposes a technique that uses the adaptive feature norm with subdomain adaptation to boost the transfer gains. Subdomain adaptation can enhance the ability of deep adaptation networks by capturing the fine-grained features from each category. Additionally, we have incorporated an adaptive feature norm approach to increase transfer gains. Our method shows state-of-the-art results on the popular visual classification datasets, including Office-31, Office Home, and Image-CLEF datasets.

Keyphrases: Domain Adaptation, Transfer Learning, object recognition

@booklet{EasyChair:6704,
  author    = {Ashiq Imran and Vassilis Athitsos},
  title     = {Adaptive Feature Norm for Unsupervised Subdomain Adaptation},
  howpublished = {EasyChair Preprint 6704},
  year      = {EasyChair, 2021}}