Leveraging multi-level regularization for efficient Domain Adaptation of Black-box Predictors

Open Webpage

Wei Li, Wenyi Zhao, Xipeng Pan, Pengcheng Zhou, Huihua Yang

Published: 01 Jan 2025, Last Modified: 14 Nov 2025Pattern Recognit. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Multi-level regularization (MLR) for domain adaptation of black-box predictors.•MLR flexibly adapts black-box model for the target domain with various architectures.•Network-level regularization via teacher-student peers for diverse representations.•Propose a mutual contrastive learning strategy to learn strong representations.•Achieves state-of-the-art performance on three datasets with fewer model parameters.