The EEG activation maps in recent work are uninterpretable by experts

Hari Bharadwaj, Ronnie Wilbur, Barak A. Pearlmutter, Jeffrey Mark Siskind

Keywords: EEG

TL;DR: We refute a central claim used to justify validity of a confounded dataset used in about one hundred papers.

Abstract:

Recent papers claim to decode object class from EEG recordings of
subjects viewing image stimuli from ImageNet and to use that
classifier to construct activation maps for the depicted object class
that are consistent with neuroscience knowledge. Empirical evaluation
of the activation maps by EEG experts calls this claim into question.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 4796