Information-theoretic regularization for Multi-source Domain Adaptation

TL;DR

This paper introduces MIAN, a novel neural network architecture that uses information-theoretic regularization to improve multi-source domain adaptation, addressing issues with multiple discriminators and outperforming existing methods.

Contribution

The paper proposes a unified domain discriminator approach with information regularization, providing theoretical insights and demonstrating superior performance in multi-source domain adaptation.

Findings

MIAN outperforms state-of-the-art methods in large-scale experiments.

Using a single discriminator simplifies the model without sacrificing accuracy.

Information regularization improves the stability and scalability of adversarial DA.

Abstract

Adversarial learning strategy has demonstrated remarkable performance in dealing with single-source Domain Adaptation (DA) problems, and it has recently been applied to Multi-source DA (MDA) problems. Although most existing MDA strategies rely on a multiple domain discriminator setting, its effect on the latent space representations has been poorly understood. Here we adopt an information-theoretic approach to identify and resolve the potential adverse effect of the multiple domain discriminators on MDA: disintegration of domain-discriminative information, limited computational scalability, and a large variance in the gradient of the loss during training. We examine the above issues by situating adversarial DA in the context of information regularization. This also provides a theoretical justification for using a single and unified domain discriminator. Based on this idea, we implement a novel neural architecture called a Multi-source Information-regularized Adaptation Networks (MIAN). Large-scale experiments demonstrate that MIAN, despite its structural simplicity, reliably and significantly outperforms other state-of-the-art methods.