Semi-supervised learning through adaptive Laplacian graph trimming

作者：

Highlights：

• A method which can adaptively fit a proper Laplacian weighted graph from data.

• A penalty helping cut inter-cluster shortcuts and enhance intra-cluster connections.

• A graph-based SSL model is less sensitive to neighborhood size by integrating ALGT.

• Superiority of ALGT is verified by experimental results on synthetic and UCI data.

摘要

•A method which can adaptively fit a proper Laplacian weighted graph from data.•A penalty helping cut inter-cluster shortcuts and enhance intra-cluster connections.•A graph-based SSL model is less sensitive to neighborhood size by integrating ALGT.•Superiority of ALGT is verified by experimental results on synthetic and UCI data.

论文关键词：Semi-supervised learning,Graph Laplacian,Self-paced learning,Nearest neighborhood graph

论文评审过程：Received 16 May 2016, Revised 5 September 2016, Accepted 16 November 2016, Available online 24 November 2016, Version of Record 22 March 2017.

论文官网地址：https://doi.org/10.1016/j.imavis.2016.11.013