- Neural networks are modeled after the way in which we understand how the human brain works. The technology aims to simulate artificial neurons, connections between them and uses some suitable mathematical functions to mimic the concept of `firing’ neurons. The training phase allows the network to correct itself in a manner that is analogous to the manner in which human beings learn. Moreover, by minimizing errors made in its prediction, we able to make the network learn by training it.
- A neural network is a function obtained by taking the composition of functions of the form v↦σ(Av+b) where A is a linear operator acting on a vector space V and b is a fixed vector b∈V. The function defined on a vector space V that maps v onto Av+b is called an affine map, and the function σ, is usually a non-linear function called an activation function (more about this later).
- In Mathematica, LinearLayer is a fully connected neural network layer that is ready to be trained. It allows to compute outputs where the size of the output is specified. Uses are able to include options for initial weights, biases and learning rate multipliers.
- In Mathematica NetTrain is a function that can be used to train a specified neural network. You will need to provide it with the training data and it minimizes the discrepancy between the given output and the network output.