Vector-derived transformation binding: an improved binding operation for deep symbol-like processing in neural networks

This repository contains the code used for the experiments and figures in the paper “Vector-derived transformation binding: an improved binding operation for deep symbol-like processing in neural networks”. It also comes with some precomputed results as running the full experiments can be quite time-intensive.

To run the code you first need a Python 3 installation (Python 2, might work, but has not been tested) with

• NumPy,
• SciPy,
• and Matplotlib

Furthermore, you will need a number of additional packages which should be installable with the command

pip install --user -r requirements.txt

The code can be run by opening the notebooks in Jupyter Notebook which can be launched with

jupyter notebook

If you just want to view the notebooks without running the code, you can also click the GitHub links to see a rendered version.