Motivated by recent development in VC-dimension of neural networks and corresponding semi-Pfaffian sets, we introduce a new method of approximating volumes and integrals for a vast class of geometric and number theoretical problems.