ABSTRACT
The neural function consists of one single layer that connects all inputs to all outputs. This learning algorithm is too simple to work beyond the simple translation and encoding of our problem; it will not even work for other encodings of characters. Modern machine learning uses an algorithm called backpropagation, which is a generalization of the basic idea of propagating the error backwards in the network over many layers, not just one. Finally, the last parameter, metrics, defines a set of metrics that should be used to measure success of the learning process. In this case, we are interested in accuracy, i.e. the distance between the true values and the predicted values during training. For this particular problem, we can generate infinite amounts of training data, because we know exactly how to implement the specified character normalization using a traditional program.
