Multi-bridge transfer learning

作者：

Highlights：

• MBTL constructs multiple latent spaces to exploit more common latent factors.

• MBTL reduces the discrepancies of the distributions in different latent spaces.

• To solve MBTL, we present an iterative algorithm with convergence guarantee.

• MBTL outperforms state-of-the-art learning methods on several datasets.

摘要

•MBTL constructs multiple latent spaces to exploit more common latent factors.•MBTL reduces the discrepancies of the distributions in different latent spaces.•To solve MBTL, we present an iterative algorithm with convergence guarantee.•MBTL outperforms state-of-the-art learning methods on several datasets.

论文关键词：Transfer learning,Non-negative matrix tri-factorization,Multi-bridge,Cross-domain classification

论文评审过程：Received 5 March 2015, Revised 8 October 2015, Accepted 12 January 2016, Available online 21 January 2016, Version of Record 20 February 2016.

论文官网地址：https://doi.org/10.1016/j.knosys.2016.01.016