Thomas Scott

An optimal control approach for neural network architecture adaptation with a posteriori error estimation

C G Krishnanunni, Thomas Scott, Tan Bui-Thanh

This work presents a novel approach for adapting neural network architecture along the depth based on a posteriori error estimation. By formulating neural network training as a continuous-time optimal control problem, we derive rigorous error estimates that quantify how approxima…Read more