Machine learning on neuroimaging data has been used to track when neural representations activate, mostly with methods such as functional magnetic resonance imaging. Researchers at the University of California Irvine set out a task and analysis pipeline built on EEG instead. Participants viewed images and words for objects from 5 categories – animals, tools, food, scenes and vehicles – and responded when two consecutive items came from the same category, with support vector machines trained within each participant to classify the recorded activity.
Both image and word trials yielded significant category classification accuracy, but image trials scored higher. In pairwise comparisons, all category pairs were statistically distinguishable for images, against a single distinguishable pair for words. At the electrode level, parietal and left temporal channels contributed more to image classification than frontal and right temporal channels. The work is a preprint and has not been through peer review.