New Neuro-Symbolic Learner Integrates OWL 2 DL Ontologies

Olga Mashkova, Asaad Mohammedsaleh, Fernando Zhapa-Camacho, Robert Hoehndorf· August 19, 2026 View original

Key takeaways

  • Baobab enables neuro-symbolic learning over complex OWL 2 DL ontologies.
  • It compiles ontologies into differentiable circuits for perception network training.
  • The system addresses limitations of prior approaches by handling full SROIQ features.
  • Baobab mitigates reasoning shortcuts, achieving Bayes-optimal posteriors in complex tasks.

Who benefits

HealthcareLife SciencesSemantic WebAI EngineeringRobotics

Summary

This paper introduces Baobab, a neuro-symbolic learner that compiles OWL 2 DL ontologies into Sentential Decision Diagrams (SDDs) to train perception networks. It addresses limitations of prior approaches by handling complex description logic features and mitigating reasoning shortcuts, achieving Bayes-optimal posteriors on real-image tasks.

Neuro-symbolic (NeSy) learning aims to combine the strengths of neural networks with symbolic reasoning. Existing NeSy approaches for description logics either sacrifice classical entailment by embedding ontologies in continuous spaces or are limited to simpler Horn fragments. This research presents Baobab, a novel system that compiles full OWL 2 DL ontologies, which are foundational for large knowledge bases in fields like biomedicine, into Sentential Decision Diagrams (SDDs). Baobab's compilation process saturates a propositional core and instantiates complex OWL 2 DL features over an active domain. The SDD's evidence-conditioned weighted model count then facilitates training a perception network to recognize real images under partial ABox supervision. The system demonstrates the ability to learn latent ontology concepts and, crucially, characterizes and mitigates "reasoning shortcuts" where independent perception collapses onto a single completion. By using a mixture indexed by query justifications, Baobab achieves Bayes-optimal posteriors, a significant advancement for non-Horn description logics.

Why it matters

For professionals working with complex knowledge bases and AI, this research offers a path to integrate rich symbolic knowledge (ontologies) directly into deep learning models, enabling more robust, explainable, and accurate reasoning, especially in domains requiring precise logical inference.

How to implement this in your domain

  1. 1Explore Baobab: Investigate the Baobab framework for integrating OWL 2 DL ontologies into neuro-symbolic AI applications.
  2. 2Develop hybrid AI systems: Design systems that leverage both neural perception and symbolic reasoning for tasks requiring both pattern recognition and logical inference.
  3. 3Enhance knowledge representation: Utilize OWL 2 DL ontologies to represent complex domain knowledge for AI systems, improving their understanding and decision-making.
  4. 4Mitigate reasoning shortcuts: Apply techniques like query-justification indexed mixtures to prevent AI models from taking unreliable reasoning shortcuts in ambiguous scenarios.

Original post by Olga Mashkova, Asaad Mohammedsaleh, Fernando Zhapa-Camacho, Robert Hoehndorf

"arXiv:2608.17741v1 Announce Type: new Abstract: OWL 2 DL ontologies, grounded in the description logic $\mathcal{SROIQ}$, express large knowledge bases in biomedicine and the Semantic Web. Neuro-symbolic (NeSy) learners over description logics either embed the ontology in a conti…"

View on X

Originally posted by Olga Mashkova, Asaad Mohammedsaleh, Fernando Zhapa-Camacho, Robert Hoehndorf on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Engineering & DevTools