Artificial Neural Networks
Artificial Neural Networks (ANN) are a mathematical representation of biological neural structures. Hereby the principal information mechanisms of natural brains and nervs are reproduced on a computer, which are used to solve complex tasks by Artificial Intelligence respectively Computational Intelligence.
Controversy to conventional analytical models, where the functionality has to be constructed to meet certain goals, ANN are trained to reproduce the desired behaviour simply by specification of various example cases, which represend the functional requirements of the model.
Hereby artificial neural nets are standing out due to some advantages:
- ANN are "universal approximators", which means that any arbitrary funtions can be expressed by the use of neural networks.
- ANN have a high degree of parallelism in the executions, which makes them very fast.
- ANN are simple in implementation. They are determined only by some matrix multiplications and some evaluations of the transfer functions, emulating the neural activation.
- ANN can easily be adopted or trained for new or modified problems without the need to change the overall solution architecture. This makes them suitable for problems, which need adaptive solution strategies.
- ANN are very general and flexible. They can be applied in the same form for different kind of tasks with an high degree in complexity.