spike-imperial/embed2sym

A neuro-symbolic reasoning system that can scale to a large number of raw inputs.

Embed2Sym

This repo contains the code for the Embed2Sym framework and experiments presented in the paper "Embed2Sym - Scalable Neuro-Symbolic Reasoning via Clustered Embeddings".

Requirements

The code was tested on Ubuntu 18.04, Python 3.8 and TensorFlow 2.7.0.

Installing

Simply clone the repo and install the dependencies appearing in the requirements.txt file.

Running the Experiments

The run.py script can be used to execute the experiments appearing in the paper.

• MNIST Addition

python run.py mnist_addition N

where N should be replaced with one of 1, 2, 3, 4 and 15.

• CIFAR-10 Addition

python run.py cifar10_addition 1

• Member

python run.py member N

where N should be replaced with one of 3, 4, 5 and 20.

• Forth Sort

python run.py forth_sort N

where N should be replaced with one of 2, 3, 4, 5, 6, 7 and 8.

Note that while the current script only runs these specific experiments, the framework is general and can easily be employed for many neuro-symbolic tasks. In the future we will add code and instructions here for running Embed2Sym on your own tasks.

Author

Embed2Sym was developed by Yaniv Aspis.
Email: [email protected]