A-NeSI: Approximate Neurosymbolic Inference

We introduce Approximate Neurosymbolic Inference (A-NeSI) a new framework for Probabilistic Neurosymbolic Learning that uses neural networks for scalable approximate inference. A-NeSI

1. performs approximate inference in polynomial time without relaxing the semantics of probabilistic logics;
2. is trained using synthetic data generated by the background knowledge;
3. can generate symbolic explanations of predictions; and
4. can guarantee the satisfaction of logical constraints at test time.

For more information, consult the papers listed below.

Requirements

A-NeSI has the following requirements:

• Python - v3.9.15
• PyTorch - v1.12.1
• TorchVision - v0.13.1
• Weights&Biases - v0.13.2

Run the following:

1. Install the dependencies inside a new virtual environment: bash setup_dependencies.sh

2. Activate the virtual environment: conda activate NRM

3. Install the A-NeSI module: pip install -e .

Experiments

The experiments are presented in the papers are available in the anesi/experiments directory. The experiments are organized with Weights&Biases. To reproduce the experiments from the paper, run

cd anesi/experiments
wandb sweep repeat/test_predict_only.yaml
wandb agent <sweep_id>

Note that you will need to update the entity and project parameters of wandb in the sweep files.

Paper

A-NeSI: A Scalable Approximate Method for Probabilistic Neurosymbolic Inference (Arxiv)

@misc{https://doi.org/10.48550/arxiv.2212.12393,
  doi = {10.48550/ARXIV.2212.12393},
  url = {https://arxiv.org/abs/2212.12393},
  author = {van Krieken, Emile and Thanapalasingam, Thiviyan and Tomczak, Jakub M. and van Harmelen, Frank and Teije, Annette ten},
  keywords = {Machine Learning (cs.LG), Artificial Intelligence (cs.AI), Logic in Computer Science (cs.LO), Machine Learning (stat.ML), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {A-NeSI: A Scalable Approximate Method for Probabilistic Neurosymbolic Inference},
  publisher = {arXiv},
  year = {2022},
  copyright = {arXiv.org perpetual, non-exclusive license}
}