Yes, it is very simple to implement a feedforward neural network using MLP/X and then load the weights into it and use it without any further training. The only area in which you would need to be careful, is that the sigmoid function and bias weights are the same in each case, as well as the structure being identical.

We recommend that:

1. You check the MLP/X network details before training the network to ensure that each network will use the same structure including sigmoid function.
2. You create a small test set of data with precisely known results and run it in each network without training and thereby ensure that the implemented network produces exactly the same results in each case.

In general, it may be easier to train the data using MLP/X, as it contains a very efficient learning algorithm.