SAID Lab

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.