Article Preview
TopMachine Learning
In the world of machine learning (Sebastiani, 2002; Michie, Spiegelhalter, & Taylor, 1994; Quinlan, 2014; Witten, Frank, Hal, & Pal, 2016; Pedregosa, et al., 2011) and/or artificial intelligence, one must find the perfect starting point and should have an idea where the project will go or possibly evolve into. There are several things we need to consider when picking a neural network design and what we want to achieve. A neural network (Kalchbrenner, Grefenstette, & Blunsom, 2014; Psaltis, Sideris, & Yamamura, 1988; Haykin, 1994; Hagan, Demuth, Beale, & De Jesús, 1996; Anthony & Bartlett, 2009) is generally comprised of 3 basic parts known as neurons, layers, and bias.
Neurons deal with numerous types of information to be processed. Each individual neuron must know how to handle these types of information: input values, weights and bias, net sum, and an activation function. Although a neuron is just a small portion, it is critical to have accurate processing of data.
Layers are important component in a neural network. There is a minimum of three layers: input, hidden, and output. Each layer must handle information being fed forward to create an expected answer. Starting with the input layer, it is critical to have information prepared accurate and normalized properly. This will lower the chance of unexpected results. The hidden layer is particularly unique, as there can be multiple depending on the complexity of processing. When more than one hidden is present, each layer will feed forward as normal and additional processing for back-propagation is acceptable. After information has traversed through the neural network layers, the output is equally important to have accurate results.