1. First clone the directory.

git submodule init; git submodule update

(If showing error of no permission, need to first add a new SSH key to your GitHub account.)

2. Install dependencies.

Create a new environment using conda, with Python >= 3.10.6 Install PyTorch (version >= 2.0.0). The repo is tested with PyTorch version of 1.10.1 and there is no guarentee that other version works. Then install other dependencies via:

pip install -r requirements.txt

The dataset files for LineWorld and LW-G can be generated using the BabyARC engine with the datasets/BabyARC submodule, the OmniGlot dataset found here, and ShapeNet can be downloaded here, the code for preprocessing the datasets are under dataset/ or directly downloaded via this link. If download from the above link, put the downloaded data under the ./datasets/files/ folder.

Download from this anonymous link: https://drive.google.com/file/d/15S1RsDy_5MdNq_iUsZCPsRHd9qMqdTBy/view?usp=share_link

Here we detail the repo's structure:

• The datasets contains the codes for loading and processing the datasets.
• The TAE.py is the implementation of model
• The SDE.py is the implementation of the diffusion model
• The attention.py is the implementation of the attention layers
• The sampler.py is a larger diffusion model based on DDPM
• The pltrain.py is a basic training script using pytorch lightning
• All files wil 3D are the 3D version of the file for handling 3D voxels.

If you find our work and/or our code useful, please cite us via:

@inproceedings{
cheng2024bridging,
title={Bridging Neural and Symbolic Representations with Transitional Dictionary Learning},
author={Junyan Cheng and Peter Chin},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=uqxBTcWRnj}
}