TILGen
Neural Network Tool

The World's Source For Fuzzy Logic Solutions

TILGen offers the following capabilities:

• Automatically generates a fuzzy expert system rulebase from a given set of system inputs and required outputs
• Simplifies the process of creating fuzzy logic expert systems
• Functions as a neural fuzzy system extractor
• Supports three learning methods:
  1. Unsupervised competitive learning (UCL)
  2. Supervised competitive learning (SCL)
  3. Differential competitive learning (DCL)
• Works in concert with other TIL software tools such as the TILShell family and Fuzzy-C Development System
• Works in conjuction with the TIL Fuzzy Programming Language (FPL), specifically designed for the implementation of fuzzy logic knowledge bases

TILGen is included with the TILShell 3.0 Professional Edition. It is available as an add-on option for other members of the TILShell family.

Overview

TILGen is a combination of two processes: a neural network that learns the input-output function described by the set of input-output data, and an algorithm that decodes the neural network to generate the rules for a fuzzy expert system.

Neural networks are trained to give a certain output based on a certain input through the use of learning methods. Generally a node of a neural network contains an activation level which, during training, is increased or decreased based on how "close" the node is to the current training input. Training methods are used to change the value of the node based on this "closeness". In general, TILGen uses the Differential Competitive Learning (DCL) technique, but other learning techniques are available.

The learning methods work by first initializing the n nodes of the neural network with the first n test vectors. Then, each remaining vector is compared with the n nodes and the node (or nodes) which most closely match the input vector is found. These nodes are commonly referred to as the winning nodes. The nodes are then updated according to the rules of the learning method:

• UCL - Rewards winning nodes by increasing their activation levels
• SCL - Rewards winning nodes by increasing their activation levels and punishes losing nodes by decreasing their activation levels
• DCL - Rewards the "closest" winning nodes and punishes the other winning nodes

The UCL and DCL techniques tend to produce similar results but DCL is guaranteed to converge quickly while UCL is not guaranteed to converge. The SCL technique generally requires more time for training and usually requires more detailed information.

Running TILGen from TILShell

TILGen is included with the Professional Edition of TILShell 3.0 and it can be activated from the Utilities pull-down menu of TILShell's Main Menu Bar. When TILGen is selected, the dialog box below will be opened:

After a rulebase has been chosen, the TILGen Options dialog box below will be opened to allow you to specify the data file to use, how to space the initial vectors, the number of winners, the cell threshold, and the name of the error file.

TIL Home Page | Ortech Engineering | TILShell | Fuzzy Logic Reservoir | Top of This Page |

This site is http://www.ortech-engr.com/fuzzy/TilGen.html
Send inquiries, comments, or suggestions to webmaster@ortech-engr.com.

Page Stats: Visitors -- Last Modified on: -- Thanks for visiting from: .