Genetic-Neural-Network

Check out for cool projects about drones and AI
Training a Neural Network using Genetic Algorithm.

Here are the results obtained after training a neural net with both backpropagation and Genetic programming.

Trained using GP (on the training set):

Trained using GP (Performance on the test set):

Trained using Backprop (on the training set):

Trained using Backprop (Performance on the test set):

Stats:

The GP approach:

Loss/Cost (on training set) after 1000 generations: 0.2231

Training Set Accuracy: 90.24%

Test set Accuracy: 91.25%

The Backpropagation approach:

Loss/Cost (on training set) after 1000 epochs: 0.2528

Training Set Accuracy: 90.00%

Test set Accuracy: 89.62%