Knowledge-inspired Subdomain Adaptation for Cross-Domain Knowledge Transfer

TL;DR

This paper introduces KISA, a novel domain adaptation framework that leverages knowledge-inspired subdomain division and fusion to improve fine-grained transfer in tasks like fraud detection and traffic prediction.

Contribution

KISA provides a new approach for subdomain division based on domain knowledge and offers theoretical insights into minimizing shared expected loss for better adaptation.

Findings

KISA outperforms existing methods on fraud detection tasks.

KISA achieves significant improvements in traffic demand prediction.

The framework demonstrates robustness across multiple domain adaptation scenarios.

Abstract

Most state-of-the-art deep domain adaptation techniques align source and target samples in a global fashion. That is, after alignment, each source sample is expected to become similar to any target sample. However, global alignment may not always be optimal or necessary in practice. For example, consider cross-domain fraud detection, where there are two types of transactions: credit and non-credit. Aligning credit and non-credit transactions separately may yield better performance than global alignment, as credit transactions are unlikely to exhibit patterns similar to non-credit transactions. To enable such fine-grained domain adaption, we propose a novel Knowledge-Inspired Subdomain Adaptation (KISA) framework. In particular, (1) We provide the theoretical insight that KISA minimizes the shared expected loss which is the premise for the success of domain adaptation methods. (2) We propose the knowledge-inspired subdomain division problem that plays a crucial role in fine-grained domain adaption. (3) We design a knowledge fusion network to exploit diverse domain knowledge. Extensive experiments demonstrate that KISA achieves remarkable results on fraud detection and traffic demand prediction tasks.