Open-Science | NINE Lab

NeurIPS 2025 EEG Foundation Challenge Now Open

Wed, 10 Sep 2025 00:00:00 +0000

The NeurIPS 2025 EEG Foundation Challenge is now open for submissions! As one of the organizers, I’m excited to invite the global research community to participate in this groundbreaking competition focused on advancing cross-task and cross-subject EEG decoding.

Challenge Overview

This challenge aims to push the boundaries of EEG analysis by developing foundation models capable of generalizing across different tasks and subjects. Participants will work with large-scale EEG datasets to create models that can:

Decode EEG signals across multiple cognitive tasks
Generalize to new subjects without extensive retraining
Handle the inherent variability in EEG data

Key Dates

Challenge Opens: September 2025
Submission Deadline: November 2025
Winners Announced: December 2025 at NeurIPS

Why Participate?

This challenge represents a unique opportunity to:

Contribute to advancing EEG analysis methodologies
Work with unprecedented large-scale EEG datasets
Compete for prizes and recognition at NeurIPS 2025
Collaborate with leading researchers in neuroscience and machine learning

Get Involved

For more information and to register, visit the official challenge website. We look forward to seeing innovative approaches that will shape the future of EEG analysis and brain-computer interfaces.

Meta Awards $130K Grant to Support Open Science at SCCN

Sat, 15 Feb 2025 00:00:00 +0000

I am excited to announce Meta’s generous gift to the Swartz Center for Computational Neuroscience. This unrestricted gift will enhance our efforts to promote open science and expand our comprehensive toolset, including EEGLAB, to encompass other modalities such as sEMG.

As co-PI on this grant, I am particularly grateful to Alexandre Gramfort for his unwavering advocacy for open science. This $130,000 award represents a significant investment in our mission to develop and maintain open-source tools for the neuroscience community.

The funding will support:

Enhancement of EEGLAB’s multimodal capabilities
Integration of sEMG (surface electromyography) analysis tools
Continued development of open science infrastructure
Training and educational resources for the global research community

This partnership with Meta underscores the importance of industry-academia collaboration in advancing scientific research and ensuring that powerful analytical tools remain freely accessible to researchers worldwide.

HBN-EEG: The FAIR implementation of the Healthy Brain Network (HBN) electroencephalography dataset

Thu, 03 Oct 2024 00:00:00 +0000

Download & Explore

📄 Paper - Full research publication
📊 HBN Dataset Portal - Access all 11 dataset releases with technical documentation
📈 Data Dashboard - Explore demographics, mental health patterns, and data quality metrics

Abstract

The Child Mind Institute (CMI) Healthy Brain Network (HBN) project has recorded phenotypic, behavioral, and neuroimaging data from ~5,000 children and young adults between the ages of 5 and 21. Here, we present “analysis-ready” data from its high-density (128-channel) electroencephalographic (EEG) recording sessions formatted as Brain Imaging Data Structure (BIDS) datasets (HBN-EEG) with behavioral and task-condition events annotated using Hierarchical Event Descriptors (HED), making it analysis-ready for many purposes, without ‘forensic’ search for unreported details. We also ensured individual data files data and event integrity and marked inconsistencies. HBN-EEG sessions include six tasks, three with no participant behavioral input and three including button press responses following task instructions. Openly available participant information includes age, gender, and four psychopathology dimensions (internalizing, externalizing, attention, and p-factor) derived from bifactor model of questionnaire data. Currently, HBN-EEG data from more than 3,000 participants is freely available on NEMAR (nemar.org) and OpenNeuro in the form of eleven dataset releases, with further dataset releases to follow. The HBN-EEG dataset is intended to support the development and validation of EEG analysis methods, including machine learning and deep learning approaches, and to facilitate the development of EEG-based biomarkers for psychiatric disorders.

The number of subjects per HBN-EEG dataset release

Citation

@ARTICLE{Shirazi2024-ye,
  title    = "{HBN}-{EEG}: The {FAIR} implementation of the Healthy Brain
              Network ({HBN}) electroencephalography dataset",
  author   = "Shirazi, Seyed Yahya and Franco, Alexandre and Hoffmann, Mauricio
              Scopel and Esper, Nathalia and Truong, Dung and Delorme, Arnaud
              and Milham, Michael and Makeig, Scott",
  journal  = "bioRxiv",
  pages    = "2024.10.03.615261",
  month    =  "3~" # oct,
  year     =  2024,
  doi      = "10.1101/2024.10.03.615261",
  language = "en"
}

The Lab Streaming Layer for Synchronized Multimodal Recording

Wed, 14 Feb 2024 00:00:00 +0000

Download

Paper
Code

Abstract

Accurately recording the interactions of humans or other organisms with their environment or other agents requires synchronized data access via multiple instruments, often running independently using different clocks. Active, hardware-mediated solutions are often infeasible or prohibitively costly to build and run across arbitrary collections of input systems. The Lab Streaming Layer (LSL) offers a software-based approach to synchronizing data streams based on per-sample time stamps and time synchronization across a common LAN. Built from the ground up for neurophysiological applications and designed for reliability, LSL offers zero-configuration functionality and accounts for network delays and jitters, making connection recovery, offset correction, and jitter compensation possible. These features ensure precise, continuous data recording, even in the face of interruptions. The LSL ecosystem has grown to support over 150 data acquisition device classes as of Feb 2024, and establishes interoperability with and among client software written in several programming languages, including C/C++, Python, MATLAB, Java, C#, JavaScript, Rust, and Julia. The resilience and versatility of LSL have made it a major data synchronization platform for multimodal human neurobehavioral recording and it is now supported by a wide range of software packages, including major stimulus presentation tools, real-time analysis packages, and brain-computer interfaces. Outside of basic science, research, and development, LSL has been used as a resilient and transparent backend in scenarios ranging from art installations to stage performances, interactive experiences, and commercial deployments. In neurobehavioral studies and other neuroscience applications, LSL facilitates the complex task of capturing organismal dynamics and environmental changes using multiple data streams at a common timebase while capturing time details for every data frame.

The overall LSL design for synchronized data recording

Citation

@ARTICLE{Kothe2024-zm,
  title    = "The Lab Streaming Layer for Synchronized Multimodal Recording",
  author   = "Kothe, Christian and Shirazi, Seyed Yahya and Stenner, Tristan and
              Medine, David and Boulay, Chadwick and Crivich, Matthew I and
              Mullen, Tim and Delorme, Arnaud and Makeig, Scott",
  journal  = "bioRxiv",
  pages    = "2024.02.13.580071",
  month    =  feb,
  year     =  2024,
  language = "en"
}