Multi-task classification with sequential instances and tasks

作者：

Highlights：

•

• We propose a new multi-task classification framework MTSIT that considers both the complexity of tasks and that of instances.

• We provide a theoretical explanation for MTSIT, which is the first to incorporate SPL into the multi-task classification based on the PAC-Bayesian theory.

• Experimental results on three real-world computer vision datasets demonstrate the effectiveness of the proposed approach.

摘要

•We propose a new multi-task classification framework MTSIT that considers both the complexity of tasks and that of instances.•We provide a theoretical explanation for MTSIT, which is the first to incorporate SPL into the multi-task classification based on the PAC-Bayesian theory.•Experimental results on three real-world computer vision datasets demonstrate the effectiveness of the proposed approach.

论文关键词：Classification,Multi-task learning,Curriculum learning,Self-paced learning

论文评审过程：Received 6 December 2017, Revised 23 February 2018, Accepted 23 February 2018, Available online 27 February 2018, Version of Record 22 March 2018.

论文官网地址：https://doi.org/10.1016/j.image.2018.02.013