ニューラルネットワークを用いた関数近似計算の精度評価

抄録

Calculation of functions on computers have evolved through the invention of approximation formulas for high-precision, high-speed calculations, leading to today's numerical calculation libraries. These libraries can compute simple functions with high precision using few operations, but the number of operations increases with function complexity. This study proposes a neural network-based function approximation method and evaluates its accuracy and operation count. The aim is to explore high-speed approximation potential, especially for complex functions with looser precision constraints.