ACDC: Online unsupervised cross-domain adaptation

作者：

Highlights：

• A novel online unsupervised cross-domain transfer.

• A fully autonomous structure that can grow and prune nodes on all training phases.

• The usage of a domain-adversarial bias–variance trade-off to adapt the discriminator.

• Domain-adversarial learning on online unsupervised cross-domain configuration.

• Source-code is made publicly available for further study.

摘要

•A novel online unsupervised cross-domain transfer.•A fully autonomous structure that can grow and prune nodes on all training phases.•The usage of a domain-adversarial bias–variance trade-off to adapt the discriminator.•Domain-adversarial learning on online unsupervised cross-domain configuration.•Source-code is made publicly available for further study.

论文关键词：Stream learning,Multistream learning,Domain adaptation,Online learning,Data streams

论文评审过程：Received 9 December 2021, Revised 6 July 2022, Accepted 16 July 2022, Available online 22 July 2022, Version of Record 4 August 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109486