Optimization methods

Location: diffice_jax/optimizer

/optimization.py

Providing provides two optimization methods:

Adam: a first-order gradient-based optimization method based on adaptive estimates of lower-order moments.

L-BFGS: Limited-memory BFGS method, a second-order quasi-Newton optimization method for solving unconstrained nonlinear optimization problems, using a limited amount of computer memory.

A stochastic training scheme is applied to two optimizers, where the code randomizes both data samples and collocation points at regular interval during training to minimize the cheating effect.