Multi-layer perceptron

MultiLayerPerceptron consists of a MATLAB class including a configurable multi-layer perceptron (or
feedforward neural network) and the methods useful for its setting and its training.

The multi-layer perceptron is fully configurable by the user through the definition of lengths and activation
functions of its successive layers as follows:
- Random initialization of weights and biases through a dedicated method,
- Setting of activation functions through method "set".

The training method of the neural network is based on the following algorithms:
- Gradient descent, with configurable learning rate, momentum and size of batches,
- Levenberg-Marquardt, with configurable parameters and an optional bayesian regularization.

The evolution of the training is viewable through an embedded visualization window and configurable in
terms of:
- Minimum mean square error (MSE),
- Number of epochs,
- Ratio between training and validation data sets.

Video demonstrations:

https://www.youtube.com/watch?v=yySd0Z4sdXo&t=0s&list=PLJXyTqQS4FL3QoWdlGj0WtT7nSYKWqBxv&index=3&ab_channel=Pseudonymeoriginal

https://www.youtube.com/watch?v=EvH7nBX-XhU&index=1&list=PLJXyTqQS4FL3QoWdlGj0WtT7nSYKWqBxv&ab_channel=Pseudonymeoriginal