eeg_disentangle

Noise-Factorized Disentangled Representation Learning for Generalizable Motor Imagery EEG Classification

Figure 9. t-SNE visualizations of latent ec extracted from the MI encoder EMI over different datasets. Blue: Training subjects, Orange: Test subjects, Circle: Left-hand, Triangle: Right-hand.

BCIC 2a BCIC 2a

OpenBMI OpenBMI

PhysioNet PhysioNet

Figure 11. Real and synthetic data trained with/without STFT losses in both time and frequency domains.

Real Real Real

Without STFT Without STFT With STFT

With STFT With STFT Without STFT

Figure 12. t-SNE visualizations of final feature layers of noise encoders with/without latent consistency loss. Different colors represent different subjects. Regarding Shapes, Circle: Lefthand, Triangle: Right-hand.

With latent consistency loss With latent consistency loss

Without latent consistency loss Without latent consistency loss