HULA
ActiveHosted by Parker Jones Neural Processing Lab

LibriBrain Word Classification

Decode which of 50 words a listener heard — from non-invasive MEG brain recordings.

Prize
£3,000
Participants
27
Teams
15
Ends in
 
Sign in to join
Join the competition to download the data

You can browse the file list now. Downloading requires accepting the rules before the entry deadline.

Sign in to join

About the data

Files

  • train/ — labelled MEG windows for the training subjects (.npz: meg (N, 306, 250) at 250 Hz, label, subject).
  • holdout/ — unlabelled test windows for each track, in canonical order.
  • sample_submission_*.csv — correctly formatted submissions with uniform probabilities.

The full-scale data lives on Hugging Face; the files here are a lightweight synthetic subset for the sandbox.

Loading

import numpy as np
d = np.load("train/subj01.npz")
X, y = d["meg"], d["label"]   # (N, 306, 250), (N,)
Sample submission
sample_submission_deep.csv

Uniform-probability submission for the Deep track. · 223 KB

Join to download
Sample submission
sample_submission_broad.csv

Uniform-probability submission for the Broad track. · 597 KB

Join to download

Files

6 files · 8.81 MB
  • sample_submission_deep.csvSample

    Uniform-probability submission for the Deep track.

    Join to download sample_submission_deep.csv
  • sample_submission_broad.csvSample

    Uniform-probability submission for the Broad track.

    Join to download sample_submission_broad.csv
holdout/2 files
  • deep.npz

    Unlabelled Deep test windows in canonical order.

    Join to download deep.npz
  • broad.npz

    Unlabelled Broad test windows in canonical order.

    Join to download broad.npz
train/2 files
  • subj00.npz

    Deep-track training windows (subject 0).

    Join to download subj00.npz
  • subj01-16.npz

    Broad-track training windows, subjects 1–16.

    Join to download subj01-16.npz