Multilayer Perceptron (MLP) for Image Classification

An example implementation of a Multilayer Perceptron in PHP 8.3, created for educational purposes.

To run the scripts, you need to install the php-gd library. It is used to parse PNG files.

sudo apt-get install php-gd

To train the model from scratch, delete the saved weights from the model folder and run the following script:

php train.php

The repository includes a pre-trained model. To verify its performance, run the command:

php test.php

The model is trained using the MNIST dataset. All external resources can be found in the References section.

References

• Mastering the Multi-Layer Perceptron (MLP) for Image Classification
• Michael A. Nielsen, "Neural Networks and Deep Learning", Determination Press, 2015
• Neural Network Course in JavaScript
• The MNIST database of handwritten digits
• Simple script to convert MNIST to PNG format
• He/Xavier initialization & activation functions: choose wisely