Does representing all of mathematics with one operator, eml(x,y)=exp(x)−ln(y), help GNNs learn symbolic reasoning? A preregistered study measured honestly at 250k-expression scale — nulls and one published gate fail included.
machine-learning reproducible-research symbolic-regression graph-neural-networks symbolic-reasoning symbolic-mathematics e-graphs mathnlp
-
Updated
Jul 31, 2026 - Python