Reproduce “EGG-SR: Embedding Symbolic Equivalence into Symbolic Regression via Equality Graph” 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
Nan Jiang, Ziyi Wang, Yexiang Xue
Researchers from the University of Texas at El Paso and Purdue University developed Egg-SR, a unified framework that embeds symbolic equivalence into modern symbolic regression algorithms via equality graphs. This approach improved accuracy for Monte Carlo Tree Search, Deep Reinforcement Learning, and Large Language Model-based methods, achieving, for instance, <1E-6 NMSE with Egg-MCTS on trigonometric datasets while maintaining stable training and minimal computational overhead.
Key insights
- Equality graphs efficiently encode symbolic equivalence, allowing for compact representation of functionally identical expressions by sharing common sub-expressions, leading to significant memory savings.
- Integrating equivalence into MCTS results in a tighter regret bound by effectively reducing the branching factor of the search tree, focusing computational effort on genuinely distinct paths.
- Aggregating rewards for equivalent expressions in DRL provides an unbiased gradient estimator with lower variance, contributing to more stable and efficient policy updates.
- The Egg module introduces negligible computational overhead to the overall symbolic regression pipeline, making its integration practical and scalable.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.