Graph variational auto-encoder for deriving EEG-based graph embedding

作者：

Highlights：

• A new decoder model for Graph Variational Auto Encoder (GVAE) is proposed.

• The proposed GVAE model derives the graph embedding of the EEG functional connection for the biometric application and is tested on 3 databases.

• The proposed method can identify users while being in steady state, being under stress, performing or imagining body movement, or stimulated by sounds, considering a few number of channels, with more than 95% accuracy.

• The GVAE model is compared with the traditional variational auto-encoder. Higher accuracy with considerably less computational cost and complexity is achieved.

摘要

•A new decoder model for Graph Variational Auto Encoder (GVAE) is proposed.•The proposed GVAE model derives the graph embedding of the EEG functional connection for the biometric application and is tested on 3 databases.•The proposed method can identify users while being in steady state, being under stress, performing or imagining body movement, or stimulated by sounds, considering a few number of channels, with more than 95% accuracy.•The GVAE model is compared with the traditional variational auto-encoder. Higher accuracy with considerably less computational cost and complexity is achieved.

论文关键词：Biometrics,Functional connectivity,Electroencephalogram (EEG),Graph variational auto encoder (GVAE),Graph deep learning

论文评审过程：Received 18 March 2021, Revised 26 June 2021, Accepted 20 July 2021, Available online 22 July 2021, Version of Record 29 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108202