Catalogue of Artificial Intelligence Techniques

   

Jump to: Top | Entry | References | Comments

View Maths as: Images | MathML

Neuro-Fuzzy systems

Aliases: Fuzzy neural networks

Keywords: fuzzy, hybrid, neural networks

Categories: Knowledge Representation


Author(s): Hsu Hlaing

A neuro-fuzzy system is one of the many hybrid systems combining fuzzy logic and neural networks. Like every intelligent techniques each of them has particular properties that are suitable for particular problems but not for the others. For example, neural networks are good at recognizing patterns, and the knowledge is automaticallly acquired by the backpropagation algorithm but the learning process is relatively slow and it is difficult to analyse the trained network. Fuzzy systems are better at reasoning with imprecise information and explaining their behaviour based on fuzzy rules thus their performance can be adjusted by changing the rules. The problem is that they cannot automaticallly acquire the rules they use for making decisions.

The neuro-fuzzy system consists of the components of a conventional fuzzy system except that computations at each stage is performed by a layer of hidden neurons and the neural network’s learning capacity is provided to enhance the system knowledge. Different architectures of neuro-fuzzy system are available. A typical fuzzy-neuro system is the ARIC architecture (approximate reasoning based intelligent control) is described by Berenji. It is a neural network model of a fuzzy controller. It learns by updating its prediction of the physical system’s behavior and adjust a predefined control knowledge base. This kind of architecture allows to combine the advantages of neural networks and fuzzy controllers. The system is able to learn, and the knowledge used within the system has the form of fuzzy IF - THEN rules. These rules are predefined therefore the system does not have to learn from scratch, resulting in a neural control system which can learn faster.ARIC consists of two coupled feed-forward neural networks, the Action state Evaluation Network (AEN) and the Action Selection Network (ASN). ASN is the multilayer neural network representation of a fuzzy conroller. AEN tries to predict the system behaviour. The ARIC architecture was applied to cart-pole balancing and it was shown that the system is able to solve this task.


References:


Comments:

Add Comment

No comments.