Multi-task and multi-kernel Gaussian process dynamical systems
作者：
Highlights：
• We propose a novel method for rectifying damaged motion sequences in an unsupervised manner.
• Our method obviates the need for complete training sequences.
• We take advantage of the sequential nature of the data, the redundancy among repetitions and across different tasks.
• We have devised efficient variational Bayesian inference.
• We have empirically evaluated on one robotic and two motion capture datasets.
摘要
•We propose a novel method for rectifying damaged motion sequences in an unsupervised manner.•Our method obviates the need for complete training sequences.•We take advantage of the sequential nature of the data, the redundancy among repetitions and across different tasks.•We have devised efficient variational Bayesian inference.•We have empirically evaluated on one robotic and two motion capture datasets.
论文关键词：Gaussian processes,Variational Bayes,Matrix decomposition,Factor models,Data completion,Human motion,Gaussian process latent variable models,Multi-task learning,Unsupervised learning
论文评审过程：Received 29 May 2016, Revised 12 December 2016, Accepted 14 December 2016, Available online 18 December 2016, Version of Record 12 March 2017.
论文官网地址：https://doi.org/10.1016/j.patcog.2016.12.014
