Baseline Genetic Programming: Symbolic Regression on Benchmarks for Sensory Evaluation Modeling

Abstract

We introduce hedonic, modelling benchmarks for thefield of sensory science evaluation. Our benchmarkframework provides a general means of defining aresponse surface which we call a 'sensory map'. Asensory map is described by a mathematical expressionwhich rationalises domain specific knowledge of theexplanatory variables and their individual or higherorder contribution to hedonic, response. The benchmarkframework supports the sensory map's so called groundtruth to be uncontrollably distorted to mimic the humanand protocol factors that obscure it. To provide abaseline for future algorithm comparison, we evaluate apublic research release of genetic programming symbolicregression algorithm on a sampling of the framework'sbenchmarks.