SymbolicArena: A Unified Infrastructure for Benchmark Distillation and Dynamic Evaluation in Symbolic Regression
SymbolicArena distills 664 symbolic-regression tasks into Core50, cutting evaluation cost 92.5% with preserved discriminability.
SymbolicArena is infrastructure for distilling and dynamically evaluating symbolic-regression benchmarks. It standardizes 664 heterogeneous tasks with executable ground-truth expressions and distills them into Core50, a 50-task benchmark that cuts evaluation workload by 92.5% while preserving coverage and algorithm discrimination. A unified execution protocol yields comparable outputs and search trajectories, and multi-axis scoring covers numerical quality, symbolic quality, and search behavior. Core50 shows 72.6% to 86.7% lower approximation error than alternative selectors, while current methods still fit numbers far better than they recover equations.
- SymbolicArena standardizes 664 symbolic-regression tasks with executable ground truth.
- Core50 distills that pool to 50 tasks and cuts workload by 92.5%.
- Distillation keeps task coverage and algorithm discriminability under balance constraints.
- Approximation error is 72.6% to 86.7% lower than alternative selectors.
- A large gap remains between numerical fitting and true symbolic recovery.
- Products
- SymbolicArenaCore50
From the original article
1 min read at arxiv.orgSymbolic regression (SR) seeks concise and interpretable mathematical expressions from data for scientific equation discovery. Existing SR benchmarks face a tradeoff between evaluation cost and benchmark validity. Repeated evaluation of large task pools is expensive, and compact benchmarks lack systematic evidence of preserved task diversity and algorithm discriminability.
By Ziwen Zhang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.