Digital circuits and neural networks based on acid-base chemistry implemented by robotic fluid handling

Introduction

This is the codebase and scripts of our paper: Digital circuits and neural networks based on acid-base chemistry implemented by robotic fluid handling.

1) Environment and Dependencies:

Tested on Linux (Debian 9.3)
Python 3.7
Uses Python's virtualenv
All needed Python packages are installable from dataset-gen/requirements.txt & simulations/requirements.txt as explained below.
Pacakge dependencies are pretty light, hence, installation should be very fast depending on the bandwidth.
Every generation or simulation script takes around 1-10 mins depending on the resources and the datasize on average desktops.

The code is split into two parts:

dataset-gen/ contains the scripts that are used to download MNIST dataset, run the model training, binarization of the weights, and the dataset.
simulations/ contains the echo scripts and the simulation files to simulate the datasets generated from the previous part.

2) Dataset:

Option A): Download the dataset:

To download the dataset files:

Navigate to simulations/
Run python download_dataset.py
The script will download and extract the dataset to simulations/simulation/datasets

Option B): Generate the dataset

It it recommended to use the dataset provided through option A, but if you prefer to generate the dataset by yourself:

Navigate to dataset-gen/
Create a Python virtual environment using: virtualenv venv
Activate the environment: source venv/bin/activate
Install the dependencies: pip install -r requirements.txt
Finally, to train the model and generate the dataset files, run: ./generate_ds_imgs.sh
This will generate the needed data under datasets/

3) Running the simulations:

To run the simulations:

Navigate to simulations/
Create a Python virtual environment using: virtualenv venv
Activate the environment: source venv/bin/activate
Install the dependencies: pip install -r requirements
Set the environment variable: export PYTHONPATH=$(pwd):$(pwd)/chemcpupy
Set the variable data_path to the corresponding datasset path.
Finally, to execute the simulations, run: ./sim_2_class_bin.sh, ./sim_3_class_bin.sh, or sim_2_class_3bit.sh to simulate the corresponding experiment from the paper.
The scripts will generate accuracy files in the form of acc_<bin or 3bit>_<number of classe>class_<image size>.txt that contains the accuracy metrics for the given simulations. The expected output should be similar to the results table in the manuscript.
To clear the generated data, run ./clean.sh to delete the generated files.

MNIST Dataset License

Yann LeCun and Corinna Cortes hold the copyright of MNIST dataset, which is a derivative work from original NIST datasets. MNIST dataset is made available under the terms of the Creative Commons Attribution-Share Alike 3.0 license.

License

See LICENSE file

Citation

If you find our work helpful in your research, please cite our paper:

@article{agiza2023digital,
  title={Digital circuits and neural networks based on acid-base chemistry implemented by robotic fluid handling},
  author={Agiza, Ahmed A and Oakley, Kady and Rosenstein, Jacob K and Rubenstein, Brenda M and Kim, Eunsuk and Riedel, Marc and Reda, Sherief},
  journal={Nature communications},
  volume={14},
  number={1},
  pages={496},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

scale-lab / AcidBaseNetworks