Classifying demonstration format and presenter identity in imitative learning task: EEG-based explainable machine...

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Imitation learning is crucial for skill acquisition.
• Understanding how the brain processes different demonstration formats (live vs. video) is vital.
• Explainable AI offers insights into neural correlates of learning.

Purpose of the Study:

• To investigate EEG signal differences between live and video gesture demonstrations.
• To explore EEG-based discrimination between individual demonstrators.
• To apply explainable machine learning for interpreting neural patterns.

Main Methods:

• Recorded EEG from 83 participants during imitation tasks.
• Extracted relative power in alpha and beta bands.
• Utilized Random Forest and Multilayer Perceptron models with Bayesian optimization and SHAP for explainability.

Main Results:

• Beta-band activity was the most informative feature for classification.
• Multilayer Perceptron models achieved up to 77% accuracy in identifying demonstration format.
• Random Forest models were more effective in distinguishing individual demonstrators.
• SHAP analysis revealed distinct neural patterns for live vs. video demonstrations, linked to social-perceptual and cognitive processing.

Conclusions:

• EEG signals, particularly beta-band activity, can differentiate gesture demonstration formats.
• Machine learning models can decode these neural differences.
• Explainable AI provides insights into the neural mechanisms underlying imitation learning from different sources.