Decision-Maker Preference Modeling in Interactive Multiobjective Optimization

This work presents a methodology for modeling the information concerning preferences which is acquired from a Decision-Maker (DM), in the course of one run of an interactive evolutionary multiobjective optimization algorithm. Specifically, the Interactive Territory Defining Evolutionary Algorithm (iTDEA) is considered here. The preference model is encoded as a Neural Network (the NN-DM) which is trained using ordinal information only, as provided by the queries to the DM. With the NN-DM model, the preference information becomes available, after the first run of the interactive evolutionary multiobjective optimization algorithm, for being used in other decision processes. The proposed methodology can be useful in those situations in which a recurrent decision process must be performed, associated to several runs of a multiobjective optimization algorithm over the same problem with different parameters in each run, assuming that the utility function is not dependent on the changing parameters. The main point raised here is: the information obtained from the DM should not be discarded, leading to a new complete interaction with the DM each time a new run of a problem of the same class is required.