Subject Conditioning for Motor Imagery Using Attention Mechanism

This paper presents an advanced approach for enhancing electroencephalography (EEG) classification accuracy in motor tasks through the integration of subject-specific features. Recognizing the significant challenge posed by inter-subject variability in EEG signal processing, our method focuses on addressing individual differences in EEG data. The proposed ‘Attentive Subject Fusion’ method leverages power spectral density characteristics to encode subject-specific information using a single-layer perceptron. Subsequently, an attention mechanism integrates these features with the actual EEG signal processed by an M-ShallowConvNet.Empirical evaluations demonstrate that incorporating subject-specific features markedly improves the performance of deep learning models in EEG motor task classification.