MLP neural network trained on the QCL [-5, +20] Å dataset

Correa Hernandez, Andres; Gmachl, Claire F.
Issue date: 20 March 2024
Cite as:
Correa Hernandez, Andres & Gmachl, Claire F. (2024). MLP neural network trained on the QCL [-5, +20] Å dataset [Data set]. Princeton University.
  author      = {Correa Hernandez, Andres and
                Gmachl, Claire F.},
  title       = {{MLP neural network trained on the QCL [-
                5, +20] Å dataset}},
  publisher   = {{Princeton University}},
  year        = 2024,
  url         = {}

This item contains two files. A multi-layer perceptron (MLP) neural network is built using the MATLAB Deep Network Designer (.m file). It imports a quantum cascade laser (QCL) dataset and splits it into 70% training, 15% validation, and 15% testing subsets. The network consists of an input layer, three hidden layers (each having a normalization and activation layer), and a regression output layer. All of the layers are fully connected, and the root-mean-square error (RMSE) is used to evaluate the accuracy of the network. An algorithm is trained on the [-5, +20] QCL dataset using 50 neurons, ReLU activation function, solver Adam, 0.001 learning rate, over 50 epochs, and is saved to be used in the prediction of figure of merit values for QCL designs (.mat file).

Show More
# Filename Description Filesize
1 README.txt 3.27 KB
2 adam_relu_hidden_3_neuron_50_epoch_50_net.mat 28.9 KB
3 mlp_reg_train.m 3.02 KB