Technologies these days have grown exponentially and every day becomes more and more necessary. I want to share valuable information about a topic where I believe I am an expert: Neural Networks.

What are neural networks?

A neural network is system that all living beings have,basically it provides intelligence and control for all biological systems. The properties that neural networks have are quite amazing compared to programmed algorithms.

A computer algorithm can perform thousans of operations per second specially with processors that are being used these days, but if any system or element of the program fails the program itself fails. Otherwise, a neural network can perform the same tasks of computer algorithm based program, but if an element or system of the network fails, the neural network it self recovers from the error. This makes neural network tolerant to errors. Also, Neural networks can perform operations that a regular linear program cannot, like image recognition, pattern recognition among others.

Because of its properties, scientists and engineers decide to implement an artificial version of a neural network, they called it Artificial Neural Networks. The architecture is the same as the biological neural network but simpler.This emulated system can do the same tasks of a biological neural system.

An artificial neural network structure is very simple. It is a group of smaller elements called neurons which each element has a set of inputs and a single output . Each input is multiplied by a weight and the value of these weights is the one that determines the output of the neuron. The result of the operation of the inputs and the weights is added together providing an output.

One disadvantage of a neural network system is that it needs to be trained in order to function compared to linear programming and some neural network architectures needs to repeat the secuence of training using all patterns. Other disadvantage is how we implementat it, because is a parallel system, it goes against the linear structure of microprocessors thus we need to emulate them.

A neural network is an useful system, unfortunally scientists were not able to use all functionalies of biological system because they are to wide and complex, so by time and research we have created different network architectures with different functionalities. The most common are:

• Backpropagation: The most popular neural network ever used. It uses the delta algorithm of training. It can be used as a data selector or a recongnition system among other applications.
• Perceptron: The first neural network, it uses a single layer. It uses the LMS algorithm.
• ART (Adaptative Resonance theory): ART is a very complex algorithm and architecture, this network has the hability of be trained without supervisor.
• Hopfield: A single layer neural network very useful as a memory. It's used to recognize information an to reconstruct if is incomplete.
• SOM: Self organizing maps - This networks is very much like than ART. Uses supervised and competitive learning method.

Recomended Site

Neural Network Theory

An introduction to artificial neural networks, very deep information about neural networks, also has information about artificial intelligence related topics like genetic algorithms, intelligent search DSP among others.