Reproduce “Categorical Representation Learning: Morphism is All You Need” 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
Artan Sheshmani, Yizhuang You
A new machine learning framework, 'categorical representation learning,' leverages category theory to explicitly learn and represent relationships (morphisms) as feature matrices or tensors. The approach achieved comparable performance in semi-supervised chemical compound translation with approximately 17 times fewer parameters than state-of-the-art sequence-to-sequence models.
Key insights
- Morphisms are not just attributes but fundamental entities that intrinsically define objects and capture the structural relations within data.
- Category theory provides a rigorous mathematical framework to unify diverse machine learning tasks by conceptualizing them as structure-preserving mappings (functors).
- Object and morphism embeddings can be learned unsupervisedly from concurrence statistics, with fuzzy morphisms providing a probabilistic interpretation of relations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.