Architecture

The architecture can be summarized as follows:

In the first stage of optimization, we explore the extent of the search space using moving average gradient descent, while in the second stage we take the best performing region as the starting point and find pseudo-optimal input parameters using the Bayesian Optimization.