Combining two emerging technologies, National Semiconductor Corp yesterday launched what it claims is the industry’s first software product that uses neural network learning to generate fuzzy logic rules and membership functions for embedded microcontroller applications. Called NeuFuz4, it runs under Windows and is claimed to reduce the cost and time needed to implement fuzzy logic because it automatically create a fuzzy system based on desired system inputs and outputs – it enables up to four inputs to create one output, then verifies and codes it for use with NatSemi’s COP8 family of embedded microcontrollers. Conventional heuristic-based fuzzy logic systems cannot learn and fail to work with many complex applications, the company says, noting that NeuFuz4 gets round the problem by using proprietary new defuzzification, rule inferencing and antecedent processing algorithms based on a modified back propagation algorithm. It uses a neural network to learn system behaviour based on system input-output data and generates fuzzy rules and membership functions based what it has learned. NeuFuz4 users provide system input-output data through the Windows interface and the neural network takes over; the modified back propagation method passes the error generated in the neural network system back to the input layer and hidden layers in a step-by-step process that helps adjust the neural network’s weights in a non-linear real world fashion. The NeuFuz4 Development System includes the Windows software, COP8 development tools, documentation, consulting and technical support for $10,000. A NeuFuz4 Development Kit with software, documentation and introductory set of COP8 development tools and support is $3,975. A Learning Kit with a limited version of the development software and documentation is $200.
