[3Rin2] Interactive Session 1

[3Rin2-04] Semi-supervised Domain Adaptation using Prediction Models in Associated Domains

〇Yasuhiro Sogawa¹, Tomoya Sakai¹ (1. NEC Corporation)

Keywords:Domain Adaptation, Semi-supervised Learning, Distillation

Semi-supervised domain adaptation which trains a prediction model so that it adapts to novel domains from a few labeled and relatively large unlabeled observations. In this talk, we consider semi-supervised domain adaptation and propose a model embedding method. Unlike the conventional semi-supervised domain adaptation, our work utilizes prediction models in source domains. Moreover, our method can generate a pseudo label to unlabeled data without any special assumption on data distribution. Through experiments, we confirm the effectiveness of our proposed model embedding approach.