Features of horrible color schemes of living things

An estimation using Gaussian process ordinal regression

Abstract

Venomous living things are known to often have characteristic color schemes called aposematism or warning coloration. Such warning colors have been known to have a role in warning predators that they are harmful and inedible. This study investigated the relationships between perceived horribleness of colored spiders and mushrooms, and the coloration parameters using a sequential experimental design with Bayesian optimization (BO) methodology. In this study, we applied BO incorporating ordinal Gaussian process regression, which enables an estimation of the latent function of horribleness based on evaluations using the Likert scale. The parameters of spider and mushroom colors were automatically calculated for each trial based on the acquisition function of UCB, which reflects all the history of the participant’s responses. Ten participants provided their evaluations of the images of spiders and mushrooms across 120 trials. Based on the results of BO, the features of coloration that the participants perceived as horrible on average were investigated using a peak detection method.