POYO (Azabou et al 2023, NeurIPS) introduces a new transformer-based framework for neural population decoding, designed to adapt rapidly to new, unseen sessions with minimal labels, leveraging large-scale neural recordings. Read here for a high-level intro to POYO.

The code for POYO has been moved to the torch_brain package. You can find the code under examples/poyo.

git clone https://github.com/neuro-galaxy/torch_brain.git
cd torch_brain/examples/poyo

To train POYO-MP you first need to download the perich_miller_population_2018 data using brainsets.

brainsets prepare perich_miller_population_2018

Then you can train POYO-MP by running:

python train.py --config-name train_poyo_mp.yaml

Checkout configs/base.yaml and configs/train_poyo_mp.yaml for all configurations available.

To train POYO-1 you first need to download all datasets using brainsets.

brainsets prepare perich_miller_population_2018
brainsets prepare churchland_shenoy_neural_2012
brainsets prepare flint_slutzky_accurate_2012
brainsets prepare odoherty_sabes_nonhuman_2017

Then you can train POYO-1 by running:

python train.py --config-name train_poyo_1.yaml

Below is a table of pre-trained weights for POYO models that you can download and use:

Please cite our paper if you use this code in your own work:

@inproceedings{
    azabou2023unified,
    title={A Unified, Scalable Framework for Neural Population Decoding},
    author={Mehdi Azabou and Vinam Arora and Venkataramana Ganesh and Ximeng Mao and Santosh Nachimuthu and Michael Mendelson and Blake Richards and Matthew Perich and Guillaume Lajoie and Eva L. Dyer},
    booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
    year={2023},
}