# Nonlinear Embedding Transform for Unsupervised Domain Adaptation

## TL;DR

This paper introduces the Nonlinear Embedding Transform (NET), a novel method for unsupervised domain adaptation that combines domain alignment with similarity-based embedding, validated through extensive vision dataset experiments.

## Contribution

The paper proposes a new nonlinear embedding transform method for unsupervised domain adaptation, including a validation procedure for parameter estimation, outperforming existing methods.

## Key findings

- NET outperforms existing unsupervised DA methods on vision datasets
- The validation procedure effectively estimates model parameters
- Comprehensive experiments demonstrate the method's robustness

## Abstract

The problem of domain adaptation (DA) deals with adapting classifier models trained on one data distribution to different data distributions. In this paper, we introduce the Nonlinear Embedding Transform (NET) for unsupervised DA by combining domain alignment along with similarity-based embedding. We also introduce a validation procedure to estimate the model parameters for the NET algorithm using the source data. Comprehensive evaluations on multiple vision datasets demonstrate that the NET algorithm outperforms existing competitive procedures for unsupervised DA.