Which of the following statements about the training of a neural network is false?

a. The training of a neural network is an iterative process of predicting an observation or batch of observations and then updating neuron weights based on a function of the prediction error.

b. The performance of a neural network does not depend on the choice of activation function used for the hidden layers.

c. To address the overfitting of a neural network, a regularization approach can be applied to penalize the magnitude of the weights placed on neurons’ output.

d. The size and number of hidden layers is determined experimentally by analyzing the impact on predictive performance in a cross-validation experiment.