PredictiveCodingBackprop

Code for the paper "Predictive Coding Approximates Backprop along Arbitrary Computation Graphs". This repo contains code to reproduce all figures and experiments in the paper. If you find this code useful please cite the paper https://arxiv.org/pdf/2006.04182.pdf

Installation and Usage

Simply git clone the repository to your home computer. The numerical_results.py file will recreate the numerical results figures in section 5.1. The cnn.py file contains the predictive coding and backprop CNNs used in section 5.2. The lstm.py and rnn_names.py files contain predictive coding and backprop LSTMs and RNNs used in section 5.3.

Requirements

The code is written in [Pyython 3.x] and uses the following packages:

• [NumPY]
• [PyTorch] version 1.3.1
• [TensorFlow] version 1.x (only for downloading shakespeare dataset)
• [matplotlib] for plotting figures

Citation

If you enjoyed the paper or found the code useful, please cite as:

@article{millidge2020predictive,
  title={Predictive Coding Approximates Backprop along Arbitrary Computation Graphs},
  author={Millidge, Beren and Tschantz, Alexander and Buckley, Christopher L},
  journal={arXiv preprint arXiv:2006.04182},
  year={2020}
}