EEG/EMG Foundation Challenge 2026#
Runnable EEGDash companions for the four announced 2026 competition tracks. Each tutorial shows the seed-corpus query, the held-out evaluation axis, and the track metric on a deterministic synthetic analogue that builds offline.
The synthetic scores are teaching checks, not leaderboard baselines. Use the official NeuralBench evaluator for competition results: https://facebookresearch.github.io/neuroai/neuralbench/auto_examples/biosignal_challenge_2026/index.html
What you will learn:
Track 1: map EEG to image embeddings and measure top-5 retrieval accuracy on unseen stimuli.
Track 2: decode BCI commands across sessions with balanced accuracy.
Track 3: predict sleep-onset latency across held-out subjects and devices, and understand the official weighted, binned error metric.
Track 4: decode EMG keystrokes across users and measure character error rate.
Run the tutorials:
tutorial_track_1_eeg_to_image.py– cross-stimulus EEG-to-image.tutorial_track_2_bci.py– cross-session BCI commands.tutorial_track_3_sleep_onset.py– cross-subject onset regression.tutorial_track_4_emg_to_text.py– cross-user EMG-to-text.
