Deep Neural Network and field experiments reveal how transparent wing windows reduce detectability in butterflies

Lepidoptera, a group of insects in which wing transparency has arisen multiple times, exhibit much variation in the size and position of transparent wing zones. However, little is known as to how this variability affects detectability. Here, we test how the size and position of transparent elements affect predation of artificial butterflies by wild birds in the field. We also test whether deep neural networks (DNNs) might be a reasonable proxy for live predators, as this would enable one to rapidly test a larger range of hypotheses than is possible with live animals. We compare our field results with results from six different DNN architectures (AlexNet, VGG-16, VGG-19, ResNet-18, SqueezeNet, and GoogLeNet). Our field experiment demonstrated the effectiveness of transparent elements touching wing borders at reducing detectability, but showed no effect of transparent element size. DNN simulations only partly matched field results, as larger transparent elements were also harder for DNNs to detect. The lack of consistency between wild predators and DNNs responses raises questions about what both experiments were effectively testing, what is perceived by each predator type, and whether DNNs can be considered to be effective models for testing hypotheses about animal perception and cognition.