BPU: Biological Processing Unit

This repository implements a novel neural network architecture that incorporates the Drosophila larva connectome (DOI: 10.1126/science.add9330) as a recurrent neural network (RNN) layer and tests its performance on the MNIST dataset.

Requirements

• Python 3.8+
• PyTorch
• NumPy
• Matplotlib
• Pandas
• Seaborn

Project Structure

BPU/
├── data/                  # Connectome data and datasets
├── notebooks/            # Jupyter notebooks for training and analysis
├── plotting/             # Visualization scripts and figures
├── results/              # Saved model weights and experiment results
├── src/                  # Source code
│   ├── connectome.py    # Connectome processing utilities
│   ├── data_loader.py   # Data loading and preprocessing
│   ├── net.py           # Neural network model definitions
│   └── utils.py         # Utility functions
├── config/              # Configuration file for experiments

Getting Started

See the tutorial notebook for a step-by-step guide on configuring experiments, loading data, training models, and visualizing results.

Citation

If you use this code or find this work helpful, please cite: