A Knowledge-Based Approach to MLP Con guration

We propose a knowledge-based approach to the task of determining the topology of mul-tilayer perceptrons (MLPs). The idea consists in integrating well-founded and empirically proven connguration techniques into a knowledge-based system. A preliminary study showed that the use of these techniques depends primarily on the amount of prior domain knowledge available: connguration techniques can be situated along a spectrum going from knowledge-intensive techniques like symbolic structure compilation, through partial techniques using pieces of domain knowledge like hints, to knowledge-lean techniques based mainly on heuristic search. Our knowledge base for MLP connguration formalizes the conditions of applicability of these techniques as well as interdependencies and complementaries among them which might lead to novel hybrid connguration strategies.