Speaker: Álvaro Sáiz López.
Abstract: P300-based brain-computer interfaces (BCIs) provide a non-muscular communication channel for patients with severe motor impairments. This work leverages BigP3BCI, a recently released dataset unifying 18 studies and ~200 subjects, to systematically compare feature extractors for P300 detection: temporal windowing, xDAWN, and EEGNet embeddings. We fine-tune a pre-trained EEGNet on BigP3BCI to produce P3EEGNet, further personalized per subject. The classical xDAWN+LDA pipeline remains a strong baseline; the fine-tuned EEGNet performs comparably but fails on different subjects, revealing a systematic complementarity. We exploit this with a subject-specific selection ensemble that matches or outperforms the best individual model, supported by statistical analysis at a scale not previously feasible in the field.