Remarks on multi-output Gaussian process regression
作者：
Highlights：
• State-of-the-art MOGPs were reviewed and analyzed.
• Ten representative MOGPs were investigated in different scenarios.
• Heterotopic training data enlarges information diversity for symmetric MOGPs.
• Symmetric MOGPs favor moderate training size, while asymmetric MOGPs favor small and moderate training sizes.
• Decomposed process helps asymmetric MOGPs perform well for complex cases.
摘要
•State-of-the-art MOGPs were reviewed and analyzed.•Ten representative MOGPs were investigated in different scenarios.•Heterotopic training data enlarges information diversity for symmetric MOGPs.•Symmetric MOGPs favor moderate training size, while asymmetric MOGPs favor small and moderate training sizes.•Decomposed process helps asymmetric MOGPs perform well for complex cases.
论文关键词：Multi-output Gaussian process,Symmetric/asymmetric MOGP,Multi-fidelity,Output correlation,Knowledge transfer
论文评审过程：Received 18 August 2017, Revised 18 October 2017, Accepted 27 December 2017, Available online 3 January 2018, Version of Record 14 February 2018.
论文官网地址：https://doi.org/10.1016/j.knosys.2017.12.034
