Reproduce “Soft Ontology Priors over Latent Concept Posteriors for Weakly Supervised Neuro-Symbolic Learning” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tengwei Song, Robert Hoehndorf

The authors introduce a weakly supervised neuro-symbolic method that adds soft ontology knowledge about hidden concepts to the training objective. Soft axioms penalize, but do not forbid, implausible concept assignments. On MNIST-Add, raw concept macro F1 is 0.498 ± 0.089 versus 0.267 ± 0.070 for HardPC, while task accuracy is similar; gains depend on a prior that resolves the task’s ambiguity.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.