docs / eeg-emg-foundation-challenge / overview

⌘ K · search Organizers GitHub Register

NeurIPS 2026 · open competition

EEG/EMG Foundation Challenge 2026.

Neural Interfaces for Generalizable Decoding is a unified, open-source benchmark for AI models of EEG/EMG activity. 94 datasets, 36 EEG tasks, 14 architectures — all behind a standardized BIDS-first interface. Pretrain once. Evaluate everywhere. Reproduce by construction.

EEG EMG Sleep BIDS-first PyTorch MIT licensed

bash · pip · neuralbench

 1# 1. Install the benchmark
 2$ pip install neuralbench eegdash braindecode
 3 
 4# 2. Download a task
 5$ neuralbench eeg eeg_to_image --download
 6 
 7# 3. Run the full grid
 8$ neuralbench eeg eeg_to_image -m eegnet
 9  ↳ NeuralBench-EEG-Core v1.0 · 94 datasets · 36 tasks

Demo signal · α-band placeholder 500 Hz

94 datasets +3 this wk

36 EEG tasks

14 architectures

12,340 h of recording

3,412 subjects

184 teams ↑ 12

Five tracks, one pipeline draft

From cognition to clinic to foundation transfer.

Each track shares the same evaluation harness, BIDS-first datasets, and reproducibility audit. Submit to one — or stack against all five.

Submissions openJul 1 — Sep 1, 2026 (AoE)

Final resultsNeurIPS · Sydney

Prize pool$XXX,000 USD

Track 01 OPEN

EEG-to-IMG

Evoked visual retrieval: rank held-out candidate images from EEG epochs recorded during visual stimulation. Targets are frozen DINOv2-giant embeddings.

Visual retrieval DINOv2 RSVP Object viewing

Trials1.6M+

Subjects80+

MetricTop-5 Acc

Current leader REVE-pretrained · Lausanne

84.7

Submit to this track → Browse datasets

Track 02 OPEN

BCI decoding

Calibration-stable command decoding across sessions and contexts. Generalize motor imagery, calculation, and word association tasks to later sessions without recalibration.

Motor imagery Mental math Word association Session drift

Sessions120+

Subjects20

MetricBal. Acc

Current leader EEGNet-baseline · Inria

58.6

Submit to this track → Browse datasets

Track 03 OPEN

Sleep onset

Estimate the latency from recording start to stable sleep onset (N2 event) using consumer-grade wearable EEG. Staging is secondary to precise timing.

Sleep onset Wearable EEG Muse Home monitoring

Sleepers1,000+

MetricMAE (s)

Current leader LaBraM-sleep · Tsinghua

134.9

Submit to this track → Browse datasets

Track 04 OPEN

EMG-to-Text

Decode typed keystrokes from wristband surface EMG. Sequence transduction across 100+ users, addressing anatomy, typing strategy, and sensor re-placement.

EMG typing Sequence transduction Wristbands QWERTY

Users108

Keystrokes5M+

MetricCER (%)

Current leader EMG2QwertyNet · Meta

25.1

Submit to this track → Browse datasets

Track 05 OPEN

Foundation transfer

The ultimate test of shift-robust representations: use a single shared EEG encoder across the Image, BCI, and Sleep tracks. Evaluates reusable cross-task learning. Fundamental models: can we do better across all tasks?

Frozen probe Finetune Shared encoder Cross-task

Tasks3

Hours10,000+

Models12

Current leader BIOT-large · Yale + ETH

82.1

Submit to this track → Browse datasets

Phases · Jul — Dec 2026

From warm-up to NeurIPS finals.

One warm-up window, one final submission window, one reproducibility audit, one weekend in Sydney. The schedule follows the NeurIPS 2026 competition guidelines: launch July 1, submissions close September 1 (AoE), final rankings November 1, present at the December workshop.

All timesAnywhere on Earth (AoE)

Daily cap (final phase)5 submissions / team / day

WorkshopNeurIPS · Dec 11–12, 2026

Phase 01 NEXT
Warm-up
Jul 1 — Jul 31, 2026
Public validation set, unlimited submissions, baseline models released on Jul 1. Get your pipeline running end-to-end before the sealed test opens.

OpensJul 1, 2026

Validation setPublic release

SubmissionsUnlimited
Phase 02 UPCOMING
Final submissions
Aug 1 — Sep 1, 2026 (AoE)
Sealed test set opens Aug 1. Daily submission caps enforced. Submissions close Sep 1 AoE. Each team submits a 2-page methods note alongside their best model.

OpensAug 1, 2026

ClosesSep 1, 2026 (AoE)

Daily cap5 / team / day
Phase 03 UPCOMING
Audit & Ranking
Oct 1 — Nov 1, 2026
Reproducibility audit of the top-ranked submissions on Oct 1. Final rankings released Nov 1; competition reports and analysis paper drafted.

AuditOct 1, 2026

Final rankingNov 1, 2026
Phase 04 UPCOMING
NeurIPS finals
Dec 11 — 12, 2026
Track winners present in Sydney at the NeurIPS Competition Track. Awards, lessons-learned keynote, and the v2.0 roadmap. Travel grants for top-3 per track.

VenueSydney · ICC

Prize pool$XXX,000 USD

Leaderboard preview placeholder data

One ranking, updated every 15 minutes during the final phase.

# Submission Track Score Δ

#01 REVE-pretrained · Lausanne T1 84.7 –

#02 REVE-BCI · Lausanne T2 68.0 –

#03 EMG2QwertyNet · Meta T4 25.1 –

#04 EEGNet-baseline · Inria T2 58.6 –

#05 REVE-sleep · Lausanne T3 134.9 –

#06 EEGNet-sleep · Inria T3 143.3 –

#07 EEGNet-visual · Inria T1 28.1 –

Next refresh in 8 min · 184 teams · 1,247 submissions Open full leaderboard →

Datasets · BIDS-first via EEGDash placeholder data

500+ public datasets, one unified interface.

Search standardized metadata, run reproducible pipelines, export model-ready features in minutes.

python · eegdash

 1from eegdash import EEGDashDataset
 2 
 3dataset = EEGDashDataset(
 4    cache_dir="./data",
 5    dataset="ds002718",
 6    subject=["012", "013"],
 7)
 8# 47 recordings · BIDS · streamed

ID Name Mod. Subj Hours Ch. SR Coverage

ds004245 THINGS-EEG · Object viewing EEG 60 40h+ 64 1000 Hz

alljoined-1.6 Alljoined-1.6M · Natural images EEG 20 160h 32 250 Hz

bci-graz Graz/BrainHero · BCI command EEG 20 24h 43 512 Hz

muse-sleep Muse Sleep-Onset · Wearable Sleep 1000 8000h 4 256 Hz

emg2qwerty emg2qwerty · EMG typing EMG 108 346h 32 2000 Hz

Showing 7 of 94 · NeuralBench-EEG-Core v1.0 Open the start-kit →

Architectures · braindecode model zoo placeholder F1

Side-by-side, on the same data, with the same protocol.

task-specific

EEGNeX

Compact CNN

F183.2

Params0.4M

task-specific

EEGConformer

CNN + Transformer

F184.1

Params1.2M

foundation

BIOT

Transformer · pretrained

F186.1

Params24M

foundation

LaBraM

Large brain model

F184.9

Params47M

foundation

CBraMod

Criss-cross transformer

F182.7

Params38M

foundation

REVE

4D positional encoding

F181.4

Params52M

foundation

SignalJEPA

Self-supervised JEPA

F180.8

Params31M

foundation

EEGPT

Pretrained universal

F179.5

Params88M

Backed by

Sponsors & organizing institutions

Compute, datasets, prizes, and engineering hours from 14 institutions across 5 countries. Yneuro hosts the platform; AWS provides the GPU cloud. Each track has its own sponsor: Alljoined for EEG-to-IMG, Meta FAIR Brain & AI for BCI decoding, InteraXon for sleep onset, and Meta Reality Labs for EMG-to-Text.

Meet the team →

Compute partner

Track sponsors

Organizing institutions

Warm-up opens Jul 1, 2026

Build with us. Benchmark with us. Publish with us.

The challenge is open-source, MIT-licensed, and community-driven. Add a task, add a dataset, add a model — every contribution makes the benchmark stronger.

bash · contribute

 1$ git clone https://github.com/neural-interfaces26/neuralbench
 2$ cd neuralbench
 3$ pip install -e ".[dev]"
 4 
 5# Add your task
 6$ python scripts/register_task.py --modality eeg

EEG/EMG Foundation Challenge 2026.

From cognition to clinic to foundation transfer.

EEG-to-IMG

BCI decoding

Sleep onset

EMG-to-Text

Foundation transfer

From warm-up to NeurIPS finals.

Warm-up

Final submissions

Audit & Ranking

NeurIPS finals

One ranking, updated every 15 minutes during the final phase.

500+ public datasets, one unified interface.

Side-by-side, on the same data, with the same protocol.

EEGNeX

EEGConformer

BIOT

LaBraM

CBraMod

REVE

SignalJEPA

EEGPT

Sponsors & organizing institutions

Build with us. Benchmark with us. Publish with us.