NeuroLog

Compositional Neurosymbolic Integration

Prototype for neurosymbolic integration implemented in PyTorch. NeuroLog outperformed in terms of accuracy and runtime all the state-of-the-art neurosymbolic engines by that time, namely DeepProbLog, ABL, and NeurASP, supporting training on datasets that those techniques do not scale. NeuroLog laid the foundations for formalizing the problems of training neural classifiers using supervision coming from logical theories and learning under imbalances in neurosymbolic settings.

Repository

NeuroLog

Relevant publications

2021

  1. AAAI
    Neural-Symbolic Integration: a Compositional Perspective
    Efthymia Tsamoura, Timothy M. Hospedales, and Loizos Michael
    In Thirty-Fifth AAAI Conference on Artificial Intelligence, February 2-9, 2021
    DOI PDF Code

2023

  1. NeurIPS
    On Learning Latent Models with Multi-Instance Weak Supervision
    Kaifu Wang, Efthymia Tsamoura, and Dan Roth
    In Proceedings of the Thirty-seventh Conference on Neural Information Processing Systems (NeurIPS), 2023
    arXiv HTML

2024

  1. arXiv
    On Characterizing and Mitigating Imbalances in Multi-Instance Partial Label Learning
    Kaifu Wang*, Efthymia Tsamoura*, and Dan Roth
    2024
    arXiv

2024

  1. Review paper
    Mapping the Neuro-Symbolic AI Landscape by Architectures: A Handbook on Augmenting Deep Learning Through Symbolic Reasoning
    Jonathan Feldstein, Paulius Dilkas, Vaishak Belle, and Efthymia Tsamoura
    2024
    arXiv