This paper presents algorithms that allow the realization ofmulti-valued functions as a multi-level network consisting of min- andmax-gates. The algorithms are based on bi-decomposition of functionintervals, a generalization of incompletely specified functions.Multi-valued derivation operators are applied to compute decompositionstructures. For validation the algorithms have been implemented in theYADE system. Results of the decomposition of functions from machinelearning applications are listed and compared to the results of anotherdecomposer
展开▼