On data classification by iterative linear partitioning
- Martin Anthony
- Discrete Applied Mathematics
We address the problem of computing and learning multivalued multithreshold perceptrons. Every ninput k-valued logic function can be implemented using a (k; s)-perceptron, for some number of thresholds s. We propose a genetic algorithm to search for an optimal (k; s)-perceptron that e ciently realizes a given multiple-valued logic function, that is to minimize the number of thresholds. Experimental results show that the genetic algorithm nd optimal solutions in most cases.