Sensory complexity and global gain in a DCNN codetermine optimal arousal state

Arousal deeply impacts behaviour and sensory processing. Whereas for an easy task, perceptual performance linearly increases with higher arousal levels, for a more challenging task, an inverted-U-shaped relationship has been described. These findings are commonly referred to as the Yerkes-Dodson law (1908). Yet, it remains unclear why perceptual performance decays with high levels of arousal in difficult but not in simple tasks. Based on recent studies linking a global gain change in sensory processing to changes in arousal state, we augmented a deep convolutional neural network with a global gain mechanism to mimic the effects of cortical arousal. With this approach, we show that the Yerkes-Dodson law can be accounted for by this global gain mechanism, acting on a hierarchical sensory system that processes stimuli of varying sensory complexity. By leveraging the full observability of our model, we reconcile conflicting findings from previous studies on sensory processing, by showing that both linear as well inverted-U-shaped gain profiles emerge in the interaction of hierarchical sensory processing and global arousal changes.