Author: Alex Clymo
Date: 24 June 2025

This repository demonstrates how to set up and train a shallow feedforward neural network in MATLAB to approximate a function $y=f(x)$ where both input $x$ and output $y$ are vector-valued, and provides custom tools for evaluating the output and Jacobian of the approximated function.

🚧 Warning! This code is very much in early development. I put it online at this early stage to encourage myself to start using Github. Please use with caution, and comments are always welcome.

1. To provide simple examples of how to define and train a neural net using MATLAB’s feedforwardnet and train functions.
2. To offer custom, vectorized functions for evaluating the output and Jacobian of the trained network by extracting the network structure into a custom netParams structure.

• main_test1.m, main_test2.m — example scripts showing how to train a network and testing the custom evaluation functions.
• netExtractParams.m — extracts the trained network’s parameters into a structure netParams.
• netEvalF.m — evaluates the output of the network for a batch of inputs.
• netEvalFp.m — evaluates the Jacobian of the network output with respect to the input for a batch of inputs.

For a complete explanation of the code structure, mathematical background, and examples, see the full pdf readme.

These codes and notes build very heavily on the work of Alessandro Villa and Vytautas Valaitis, whose paper "A Machine Learning Projection Method for Macro-finance Models" (QE, 2024) is a fantastic reference for explaining the basics of neural nets and machine learning to a macroeconomist. My codes build on the codes they made available at their Github repository here.

• MATLAB (tested on R2024b) with Optimization and Deep Learning toolboxes