Learning from label proportions for a visual matrix speller (ERP)
EEG recordings from 13 healthy participants performing an online copy-spelling task with a visual 6x7 matrix ERP speller. The speller was designed for learning from label proportions (LLP), an unsupervised, calibration-free classification approach for ERP-based brain-computer interfaces. Two stimulus sequences with different target/non-target ratios allowed the classifier to reconstruct mean target and non-target responses without labeled data. Participants spelled a German pangram three times, with 84.5% of characters spelled correctly on average without prior calibration.
AI-generated description, may include mistakesLoading demographics…
Coming soon. Per-file data-quality summaries are precomputed by the NEMAR processing pipeline. The static aggregate is on the way — tracked at nemar-cli#511.