Sensory cheating: adversarial body patterns can fool a convolutional visual system during signaling

Abstract

Animals often assess each other by paying special attention to signals, which help to communicate the quality of each individual. When there is a conflict of interest between the signaler and the receiver, then the signaler has an incentive to cheat by producing signals which exaggerate its apparent quality. One opportunity for cheating might be to rely on sensory illusions, but it has been difficult to study sensory cheating because we have lacked quantitative models of complex visual perception. Here we address this problem by taking advantage of recent advances in modeling visual brain areas as convolutional neural networks. Given these models, we use the technique of adversarial perturbations to show how sensory cheating can shape animal appearance while nevertheless resulting in an evolutionarily stable signaling system. In our simulations, animals typically evolve exaggerated color patterns which might be analogous to the evolution of colorful body patterns in guppies.