From: Generative adversarial networks in EEG analysis: an overview
Study | Purpose | Dataset | GAN type | Evaluation metrics | Results (with GAN) |
---|---|---|---|---|---|
Ming et al., 2019 [46] | Overcome challenges for Bio-signals as intra- and cross-subject variance | MNIST  +  private dataset (driving) | SAN | Classification accuracy |  + 1% |
Panwar et al., 2019 [48] | Address training instability and frequency artifacts | BCIT X2 | cWGAN-GP | Classifier AUC | • Same subject evaluation + 3.28% • Cross-subject evaluation + 5.18% |
Panwar et al., 2020 [45] | Generate EEG data to improve the classification performance of cognitive events | BCIT X2 | WGAN-GP  +  CC-WGAN-GP | Classifier AUC | CC-WGAN-GP: + 5.83% |
Kunanbayev et al., 2021 [42] | Overcome the scarcity problem of training for robust classifier model | P. Arico et al. | DCGAN  +  WGAN-GP | Classification accuracy |  + 2%: + 4% |