Exhaustive Canonical Skeletons Generation: A Formal Dataset for Decomposable Symbolic Regression

Key points

• To address this, we introduce an exhaustive dataset of univariate canonical skeletons generated via a deterministic constrained context-free grammar.1
• A canonical skeleton acts as a generic structural representation that subsumes more specific functional forms, eliminating redundant expressions and avoiding repeated computational efforts.1
• Although our dataset consists of univariate skeletons, it is designed to support the decomposable SR paradigm to solve multivariate SR problems without assuming linear separability.1
• Canonicalizing two existing neural SR datasets reveals that our dataset is substantially more compact, eliminates structural repetitions, and achieves a uniform operator distribution free of simplicity bias.1

Sources (1)

• [1]Exhaustive Canonical Skeletons Generation: A Formal Dataset for Decomposable Symbolic Regression
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 12:39 PM

  Neural symbolic regression (SR) models require diverse datasets of mathematical expressions for robust training.

  To address this, we introduce an exhaustive dataset of univariate canonical skeletons generated via a deterministic constrained context-free grammar.

Extractive summary: sentences quoted from the sources.