Noah WilliamsParticipant
Nice — did you z-score per channel or per epoch?
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
All threads
Baseline: EEGNet gets ~0.27 BalAcc@10 on Deep
Jonas LindqvistParticipant
Sharing a quick baseline in case it helps anyone get started:
model = EEGNet(n_chans=306, n_classes=50)
# 30 epochs, AdamW, lr=3e-4
Public BalAcc@10 ≈ 0.27. Notebook link in my profile.
4 replies
Per channel, using training-set statistics.
- Ana SousaParticipant
Adding subject embeddings gave me another +0.03 on Broad.
- Zainab HassanParticipant
Confirming similar numbers. Thanks for sharing!
Sign in to reply and upvote.Sign in