AM-MTEEG: multi-task EEG classification based on impulsive associative memory.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalogram-based brain-computer interfaces (BCIs) face challenges due to cross-subject variability and limited data.
• Existing methods struggle to generalize effectively across different individuals.

Purpose of the Study:

• To propose a multi-task (MT) classification model, AM-MTEEG, for robust cross-subject EEG classification.
• To address limitations in current BCI technology by leveraging shared features and individual-specific classification.

Main Methods:

• Developed AM-MTEEG, a deep learning model combining convolutional networks, impulsive neurons, and bidirectional associative memory (AM).
• Treated each subject's EEG classification as an independent task within a multi-task framework.
• Extracted shared features across subjects using a convolutional encoder-decoder and impulsive neurons.
• Employed a Hebbian-learned AM matrix for within-subject EEG classification.

Main Results:

• AM-MTEEG demonstrated improved average accuracy compared to state-of-the-art methods on two BCI competition datasets.
• The model significantly reduced performance variance across subjects.
• Visualization revealed a precise mapping between neuronal impulses and specific movements, indicating biological interpretability.

Conclusions:

• AM-MTEEG offers a promising solution for cross-subject EEG classification in BCI applications.
• The model enhances BCI performance by effectively handling inter-subject variability and providing interpretable results.