The 2026 PNPL Competition: Word Classification and Efficient Cross-Subject Generalisation in LibriBrain100
The competition shifts non-invasive speech decoding from phonemes toward words, with a hard test on adapting to new subjects from minutes of MEG data.
The 2026 PNPL setup uses LibriBrain100, adding 32 subjects with about 40 minutes each alongside roughly 80 hours of within-subject data. It defines a Deep track for best possible within-subject word classification and a Broad track for cross-subject generalisation. The Broad track cuts subject-specific fine-tuning data from about 40 minutes to 20 and then 10, which the authors frame as clinically feasible for future BCI work. HF Daily Papers' note
The 2026 PNPL setup uses LibriBrain100, adding 32 subjects with about 40 minutes each alongside roughly 80 hours of within-subject data. It defines a Deep track for best possible within-subject word classification and a Broad track for cross-subject generalisation. The Broad track cuts subject-specific fine-tuning data from about 40 minutes to 20 and then 10, which the authors frame as clinically feasible for future BCI work. HF Daily Papers' note
score 4