The first goal of this project is to code a library able to make and run a perceptron neural network with any number of hidden layers. Activation function should be parametizable.
Two running modes are expected :
- one to find the best hyper-parameters of a neural network for a case : with training, test and validation sets
- the other more standard only with training and test sets
In a second time, it will be about changing the rules :
- using different kind of hidden layers
- disactivating some links between nodes
- introducing genetic algorithm to help the network to learn better
- ...