Choose a track and build a model.
Built on open source
One ecosystem. One path to the leaderboard.
The challenge is built on an open-source stack for neurophysiology data, models, reproducible experiments, and official evaluation. See what each independent project contributes and how they connect.
Connected tools · one shared path
How the open-source stack connects.
02
Open development workbench: choose how to build and train.
Path A
NeuralBench and/or Benchopt
Use either tool independently, or combine them.
Defines the public tasks, datasets, preprocessing, splits, and reference baselines. Use its neuro-specific environment to run a catalogue model — a task-specific network or a pretrained foundation model (probed or fine-tuned).
Open the challenge start kits ↗
Shared task & data layer
same tasks, splits & metrics in both
Benchopt
Reproducible experiment framework
Run reproducible comparisons using the provided baselines or your own model architecture. For this challenge, Benchopt reuses the NeuralBench task and data layer to train, test, validate, and package your model against the submission contract.
Open the competition kit ↗
Ready integrations and reusable resources
Open foundations
Models and public data available to both paths.
Use the same resources directly in your code
Path B
Your own pipeline
Independent from NeuralBench and Benchopt.
Develop in your preferred codebase
Review track requirements →
Train with permitted data. Only the track inputs, outputs, and final package contract must match.
submission.py
trained weights
Official evaluation
- Phase data are mounted
- Your model runs on a competition worker
- The official metric is computed
Final destinationLeaderboard score
Independent projects · connected here
Explore the open-source stack.
Each project serves a broader research community beyond this competition. Together, they provide the data, models, benchmarks, and evaluation infrastructure used here.
NeuralBench
A unified framework for benchmarking task-specific and foundation models across neurophysiology tasks.
Visit NeuralBench ↗
Reproducible experiments
Benchopt
A framework for transparent, reproducible, and comparable machine-learning and optimization benchmarks.
Visit Benchopt ↗
Benchmark platform
Codabench
An open-source platform for hosting customizable benchmarks, code submissions, scoring, and leaderboards.
Visit Codabench ↗
Deep-learning models
braindecode
A PyTorch-native toolbox for end-to-end decoding of EEG, ECoG, MEG, and other physiological signals.
Visit braindecode ↗
BCI benchmarking
MOABB
Standardized, reproducible comparisons of EEG-based BCI pipelines across public datasets and evaluation schemes.
Visit MOABB ↗
Dataset discovery
EEGDash
A Python library for finding and working with 700+ BIDS-first public neurophysiology datasets.
Visit EEGDash ↗
Open data archive
NEMAR
An open resource for finding, assessing, processing, and downloading BIDS-formatted neurophysiology data.
Visit NEMAR ↗Use it, extend it, cite it.
Every component above is independently maintained and open to inspection. Reuse what helps, extend what you need, and credit the projects that support your work.