Meeting Abstract
Sensory bias plays a pivotal role our modern theory of sexual selection. Despite this, mechanisms for how biases emerge are only partially understood. The efficient coding hypothesis posits that organisms’ sensory systems evolve to represent environmental stimuli in a way which is the least metabolically or developmentally costly. One way this can be accomplished is by removing statistical redundancies from sensory inputs to minimize the number of spikes required to represent a stimulus. In the context of visual perception, the visual cortex is thought to perform these computations. Since visual statistics vary by habitat type, the efficient coding hypothesis would predict that species will evolve to most efficiently encode their species-specific visual habitat. This is potential route for the evolution of a sensory bias; many studies in humans and non-humans have shown preferences for efficiently coded stimuli. If males can mimic the visual statistics of their environment, it could increase their attractiveness to females. We aim to test this prediction, using darters (genus Etheostoma) as a model system. We collected males representing four species with distinct habitat preferences, and captured photographs of males themselves and their native habitats. By training a sparse autoencoder on each habitat type, we can roughly mimic how the visual cortex processes information. By inputing images of males into networks trained on their habitat type, we can see whether the network has a lower activation level for males from that habitat versus males from other habitats. The lower activation level of the network is analogous to the minimization in spikes predicted by the efficient coding hypothesis. This would also represent compelling evidence that male displays have evolved to match the visual statistics of their environments through sexual selection.
