> neural program induction in compilers to try generate faster implementations Can u provide more details? I am familiar with garden-variety ML-optimization-o

> neural program induction in compilers to try generate faster implementations Can u provide more details? I am familiar with garden-variety ML-optimization-of-combinatorical-space in compilers. How would u even "measure" the expected performance unless these are gradual greedy changes