This repository is the official Mosaic code repository provided by Yigit Demirag.

Getting Started

To install the dependencies, run:

conda env create -f mosaic-env.yml
conda activate mosaic-env

To train neuromorphic Mosaic architecture (with 8x8 distributed neuron tiles with 32 neurons on each tile) on Spiking Heidelberg Digits (SHD) dataset, run:

python train.py --lambda_con=0.1 --lr=0.001 --n_core=64 --n_epochs=500 --n_rec=2048 --noise_std=0.05 --noise_str=0

This will train a particular spiking recurrent neural network architecture for 500 epochs and exports the hardware-aware small-world connectivity matrix connectivity_matrix.png during the training.

PS: We used seeds {42,52,62,72,82} for our experiments.

Acknowledgement

Our preprocessing of Spiking Heidelberg Digits (SHD) and Spiking Google Speech Commands (SSC) datasets is based on the code from SpyTorch from Friedemann Zenke.

Citation

If you find this code useful, feel free to cite our paper:

@article{Dalgaty2024,
  title = {Mosaic: in-memory computing and routing for small-world spike-based neuromorphic systems},
  volume = {15},
  ISSN = {2041-1723},
  url = {http://dx.doi.org/10.1038/s41467-023-44365-x},
  DOI = {10.1038/s41467-023-44365-x},
  number = {1},
  journal = {Nature Communications},
  publisher = {Springer Science and Business Media LLC},
  author = {Dalgaty,  Thomas and Moro,  Filippo and Demirağ,  Yiğit and De Pra,  Alessio and Indiveri,  Giacomo and Vianello,  Elisa and Payvand,  Melika},
  year = {2024},
  month = jan 
}

License

MIT