Individual-to-Individual EEG Conversion Using Swin Transformer

Краткое изложение: In cognitive and computational neuroscience, the challenge of generalizing EEG-based models across subjects stems from inter-individual variability and low signal-to-noise ratios (SNR). Traditional EEG analysis methods often fail to account for these differences, resulting in models that perform well on training data but poorly on new subjects. To address this, we propose SwinEEG, a novel individual-to-individual EEG conversion framework inspired by the Swin Transformer, a vision transformer known for its hierarchical architecture and self-attention mechanisms. SwinEEG leverages contrastive learning and image reconstruction to map neural representations between subjects by incorporating both image stimuli and source-subject EEG data. Experiments on the THINGS EEG2 dataset demonstrate improved inter-subject neural mapping performance, with an average conversion accuracy increase from 71.2 to 74.3 compared to prior methods. This approach highlights the potential of combining vision-inspired transformers and contrastive learning to mitigate individual differences and noise in EEG signals.