Understanding how the human brain works is one of the key challenges that science and society face. The Algonauts Challenge proposes a test of how well computational models do today. This test is intrinsically open and quantitative. This will allow us to precisely assess the progress in explaining the human brain.
Our experience of the world is one of objects and persons moving continuously and interacting in events. The primary target of the 2021 challenge is predicting human brain responses to short video clips of such everyday events.
Watching events unfold over time activates large swathes of the human cortex. We pose the question: How well does your computational model account for that response?
The main goal of the Algonauts Project 2021 Challenge is to use computational models to predict brain responses recorded while participants viewed short video clips of everyday events.
We provide functional MRI data collected from 10 human subjects that watched over 1,000 short video clips.
Please click here to learn more about the fMRI brain mapping and analysis that we conducted.
There are 2 challenge tracks: the Mini Track and the Full Track. The Mini Track focuses on pre-specified regions of the visual brain known to play a key role in visual processing. The Full Track considers responses across the whole brain. Participants can play in the Mini Track, Full Track, or both.
The task is: Given a) the set of videos of everyday events and b) the corresponding brain responses recorded while human participants viewed those videos, use computational models to predict brain responses for videos for which we held out brain data.
The goal of the Mini Track is to predict brain responses in specific regions of interest (ROIs), that is pre-specified regions of the brain known to play a key role in visual perception. Participants submit the predicted brain responses for each ROI in the format described in the development kit. We score the submission by measuring the predictivity for each voxel in all the ROIs for all the subjects and show the overall mean predictivity in the leaderboard calculated across voxels, ROIs, and subjects.
We provide the following data for the Mini Track (download here):
|Rank||Team Name||Challenge Score||V1||V2||V3||V4||LOC||EBA||FFA||STS||PPA|
Each brain region is scored based on their noise-normalized correlation with our held-out brain data. The challenge
score is the average noise-normalized correlation across all 9 brain regions.
Table Last Updated: 7/29/2021 (Live Updates Here)
The goal of the Full Track is to predict brain responses across the whole brain. Participants submit predicted whole-brain responses (for the provided set of reliable voxels) in the format described in the development kit. We score the submission by measuring the predictivity for each voxel in the selected set for all the subjects and display the overall mean predictivity in the leaderboard calculated across all voxels and all subjects.
We provide the following data for the Full Track (download here):
|Rank||Team Name||Challenge Score|
The challenge score is the average noise-normalized correlation with our held-out brain data across all voxels.
Table Last Updated: 7/29/2021 (Live Updates Here)
To determine how well your models can predict brain responses we compare your submitted synthetic brain data (i.e. those predicted from your model to the left out video clips) to the empirically measured brain responses. The comparison is carried out using Pearson's correlation, comparing for each voxel the 102-dimensional vector formed by the activations for the 102 test set video clips.
There are different ways to predict brain data using computational models. We put close to no restrictions on how
you do so (see Challenge Rules). However, a commonly used approach is to use voxel-wise encoding models, and we
provide a development kit to implement such a model.
Click here to learn more about voxel-wise encoding
Please click here to download the data for the Algonauts Project 2021 Challenge. The link to the Development Kit on GitHub is below.
The training data consist of 1,000 different 3-second video clips, plus the fMRI brain responses to them.
The test data consist of 102 different 3-second video clips only, and we keep the fMRI data held back.
We provide a development kit here on GitHub to help you get started. It consists of Python scripts to:
1. Participants can use any (external) data for model building, any model, and any procedure to predict brain data with one exception: Participants that use the test set for training will be disqualified (in particular brain data generated using the test set).
2. Each participant (single researchers or team) can make 1 submission per day per track.
3. Participants must upload a short report (~4-8 pages) describing their model
building process for their best model to a preprint server (e.g. bioRxiv, arXiv), or
send a PDF by email to email@example.com.
Please use this form to submit the challenge report within 7 days of the challenge submission deadline to be considered for the evaluation of challenge outcomes. Participants that do not make their approach open and transparent cannot be considered. We additionally encourage participants to make their code available online (e.g. on Github) and link to this during their form submission.
The experiments were conducted at the Athinoula A. Martinos Imaging Center at the McGovern Institute for Brain Research, Massachusetts Institute of Technology, on a Siemens PrismaFit 3T scanner (Erlangen, Germany) that was supported with funding from a NIH Shared Instrumentation Grant Program; specifically, grant number 1S10OD021569.
|Training data, test data, and development kit released:||May 1st, 2021|
|Challenge submission deadline:||August 15th, 2021 at 11:59pm (UTC-4)|
|Challenge report submission deadline:||August 22nd, 2021|
|Challenge results released:||August 23rd, 2021|
|Session at CCN 2021 (virtual):||September 7th, 2021|