Transfer Learning vs. Domain Adaptation | one minute introduction

Are the terms transferable?

Pan, Sinno Jialin, and Qiang Yang. “A survey on transfer learning.” IEEE Transactions on knowledge and data engineering 22.10 (2009): 1345–1359.

This summary is not based on a specific paper, but rather is just an introduction to the differences between the concepts of transfer learning and domain adaptation in machine learning

Prerequisite knowledge: Domains and tasks

1. Why? Oftentimes we don’t have enough data to train a deep learning model for a problem, but we can use transfer learning or domain adaptation strategies to adapt a model from a different but related task. But these terms are often used interchangeably, which is confusing.
2. What? Transfer learning is a more general term, but domain adaptation is a specific case where Yˢ = Yᵗ (see below).
3. How? Given a source domain (Dˢ)and a somewhat related target domain (Dᵗ) (with their respective tasks), transfer learning is the basic idea of exploiting the related information in {Dˢ, Tˢ} to learn P(Yᵗ|Xᵗ), while domain adaptation (in a subset of transfer learning called transductive transfer learning) is specifically for cases where the source and target tasks are the same: Yˢ = Yᵗ and P(Y|Xˢ) = P(Y|Xᵗ)(although the second assumption is often dropped, leaving just Yˢ = Yᵗ).