Few-shot activity learning by dual Markov logic networks
作者：
Highlights：
• The method adopts double expansion in MLN to avoid the effect of potential function.
• Two MLN models are used to judge the final calibration results jointly.
• When predictions diverge, dual-model cross-validation is used to determine results.
• The method applies to MC-SAT and Gibbs sampling under appropriate iteration times.
• The method has advantages in calibration accuracy and time efficiency.
摘要
•The method adopts double expansion in MLN to avoid the effect of potential function.•Two MLN models are used to judge the final calibration results jointly.•When predictions diverge, dual-model cross-validation is used to determine results.•The method applies to MC-SAT and Gibbs sampling under appropriate iteration times.•The method has advantages in calibration accuracy and time efficiency.
论文关键词：Few-shot learning,Markov logic network,Unlabeled data calibration,Dual-model cross-validation,Least square method
论文评审过程：Received 29 April 2021, Revised 8 December 2021, Accepted 3 January 2022, Available online 10 January 2022, Version of Record 2 February 2022.
论文官网地址：https://doi.org/10.1016/j.knosys.2022.108158
