Sorry for the formatting, I'm a little rushed, but here are some tips.
"In the first place, the paradigm of neurocomputation is motivated by
cognitive abilities of biological neural networks, inspired by the
knowledge of neuroscience. The foundations of ANNs are the model of
a neuron, network topology, and learning paradigms. O
model of McCulloch and Pitts serves as the basic model of a neuron
artificial.
Artificial Neuraus Networks are very useful in expert systems,
that is, in replicate and repeat procedures. "¹ (RAUBER, T. W.)
To begin, I suggest trying to understand a little bit about how the human brain works and how cognitive analysis works. It would be interesting also mathematical and physical knowledge because many equations are complex.
One of the preconceived languages for AI is Prolog
that can help with predicative programming. According to Wikipedia:
"Prolog is a programming language that fits the paradigm of
Programming in Mathematical Logic. "
NOTE: Remember that Neural Network is a expert architecture! Applying it in an environment with multiple functionality can work and the use of Fuzzy
logic is indicated.
"RNA has numerous algorithms for pattern recognition:
Kohonen, Perceptron, Adaline, Backpropagation and many others, each
with its specificity. The main advantage of using the
Backpropagation is that it works with multilayers and resolves
"non-linearly separable" problems and some algorithms do not
solve.
In short, a "non-linearly separable" problem is one where
we can separate two distinct classes in the two-dimensional Cartesian axis
just drawing a straight line. "² (DEV Media)
A widely used form of the neural network is the Backpropagation Backpropagation - DEV Media
Here is a link to an introductory presentation found on the internet, sorry if you already know the same.
Presentation RNA
¹-RAUBER, T. W. Artificial neural networks. Federal University of Espírito Santo, 2005.
² - http://www.devmedia.com.br/redes-neurais-artificiais-algoritmo-backpropagation/28559
Accessed on 06/03/2016