Researchers have developed Moose, a novel neuro-symbolic learning method designed for the OWL 2 EL profile, which is utilized in large-scale ontologies like Gene Ontology and SNOMED CT. Unlike previous methods that handle propositional theories or Datalog, Moose addresses reasoning-shortcut awareness within ontology settings. The method compiles OWL EL TBoxes and ABoxes into Sentential Decision Diagrams, enabling differentiable weighted-model-counting and overcoming expressivity limitations with closure clauses for partial supervision. Moose demonstrates improved performance over existing baselines on tasks involving latent concept learning, as evaluated on MNIST-with-ontology and pizzaiolo datasets, and provides the first reasoning-shortcut analysis in an OWL EL context. AI
IMPACT This research advances neuro-symbolic AI by enabling more sophisticated concept learning and reasoning within complex ontological structures.
RANK_REASON The cluster describes a new method presented in an academic paper for a specific type of ontology learning.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
