Sequential memory with temporal predictive coding

1. Description

This repository contains code to perform experiments with temporal predictive coding in sequence memory tasks, which is discussed in the NeurIPS 2023 paper Sequential Memory with Temporal Predictive Coding.

2. Installation

To run the code, you should first install Anaconda or Miniconda (preferably the latter), and then clone this repository to your local machine.

Once these are installed and cloned, you can simply use the appropriate .yml file to create a conda environment. For Ubuntu or Mac OS, open a terminal, go to the repository directory; for Windows, open the Anaconda Prompt, and then enter:

1. conda env create -f environment.yml
2. conda activate seqmemenv
3. pip install -e .

3. Use

Once the above are done, you can simply run a script by entering for example:

python multilayer.py

A directory named results will the be created to store all the data and figures collected from the experiments.

4. Citation

For those who find our work useful, here is how you can cite it:

@inproceedings{NEURIPS2023_8a8b9c7f,
 author = {Tang, Mufeng and Barron, Helen and Bogacz, Rafal},
 booktitle = {Advances in Neural Information Processing Systems},
 editor = {A. Oh and T. Neumann and A. Globerson and K. Saenko and M. Hardt and S. Levine},
 pages = {44341--44355},
 publisher = {Curran Associates, Inc.},
 title = {Sequential Memory with Temporal Predictive Coding},
 url = {https://proceedings.neurips.cc/paper_files/paper/2023/file/8a8b9c7f979e8819a7986b3ef825c08a-Paper-Conference.pdf},
 volume = {36},
 year = {2023}
}

5. Contact

For any inquiries or questions regarding the project, please feel free to contact Mufeng Tang at mufeng.tang@ndcn.ox.ac.uk.