am-mteeg-classification

Cross-subject EEG classification can be improved using an associative-memory-based multi-task learning framework. This hippocampus-inspired approach harmonizes inter-subject variability by sharing features across subjects while applying multitask adapters for subject-specific nuances.

Cross-subject EEG classification requires preprocessed PyTorch tensors of shape (batch, channels, timepoints). You must also define the specific number of classes and tasks before feeding the EEG data into the spiking neural network and associative memory pipeline.

Spiking neural networks can be applied to diverse BCI tasks including motor imagery, emotion recognition, and attention monitoring. The associative memory framework handles variable numbers of EEG channels and timepoints across these different multitask learning scenarios.

Associative memory improves cross-subject BCI generalization by using spiking-like encoding to create interpretable waveform representations. This hippocampus-inspired mechanism maps shared features across subjects, reducing the need for extensive individual calibration.

PyTorch and basic neural network utilities are required to run this hippocampus-inspired multi-task learning on EEG data. The framework relies on these dependencies to build the end-to-end pipeline of encoding, spiking representation, and associative memory classification.