When considering using neural networks to model data, you have to consider the advantages and disadvantages. While our model has converged faster than previous models, and perhaps with greater accuracy, this comes with a price; we are training many more model variables and there is a greater chance of overfitting. To check if overfitting is occurring, we look at the accuracy of the test and train sets. If the accuracy of the training set continues to increase while the accuracy on the test set stays the same or even decreases slightly, we can assume over-fitting is occurring.
To combat underfitting, we can increase our model depth or train the model for more iterations. To address overfitting, we can add more data or add regularization ...
