The goal of training neural networks (NNs) is not to learn an exact representation of the training data itself, but rather to build a statistical model of the process which generates the data.
• In practical applications of a feedforward NN, if the network is over-fit to the noise on the training data, especially for the small-number training samples case, it will memorize training data and give poor generalization.
• Controlling an appropriate complexity of the network can improve generalization.
Get Answers For Free
Most questions answered within 1 hours.