Someone presents to you a multilayer perceptron (MLP) model with 6 inputs and 1 output (illustrated below). All activation functions used are RELU (a typical choice for deep learning models). hidden layers output layer input layer This MLP model is used to predict the values of ores containing mineral X. Specifically, given a piece of ore (a piece of rock) containing mineral X, the 6 inputs are 6 different features (weight, size, etc.), the output is the predicted value of this piece of ore. Every feature is a continuous value bounded below and above by two known bounds. The model has been well trained (i.e., trained using many samples and it seems to work well). However, you are worried that one day it may give you a "surprise". For example, maybe for some feature combination, the predicted value could be 1 trillion (some unreasonably high value). How can you figure out the maximum value this model will produce?