Rohit Ravi

MEG

A new dimension to access content of this website

Step plot of every leaderboard submission from 8 to 16 October 2026: the Broad track rises from 0.20 to 0.722 and the Deep track from 0.57 to 0.729, with the MEG-XL reference at 0.430 and the d'Ascoli et al. reference at 0.732 marked

LibriBrain 2026: decoding speech from MEG

MEG word decoding for LibriBrain 2026, with public scores of 0.722 on Broad and 0.729 on Deep. Includes transfer learning, test-time adaptation, and an evaluation of timing-only baselines.

Building my LibriBrain speech decoder

How I developed the LibriBrain entry: changing the training target, transferring across listeners, testing sentence-level context, and evaluating shortcuts under competition constraints.

 Every LibriBrain 2026 leaderboard submission from 8 to 16 October, with the best Broad score rising from 0.20 to 0.722 and the best Deep score to 0.729

What timing reveals about MEG decoding benchmarks

Technical notes from LibriBrain 2026: contextual training targets, cross-subject transfer, test-time adaptation, and a timing-only baseline that changes how decoder scores should be interpreted.