Walker, E. Y. et al. Nat. Neurosci. 22, 2060–2065 (2019).

Identifying the sensory stimuli that optimally excite neurons in sensory pathways is important for understanding information processing. However, the stimulus space in, for example, the visual system is vast, and identifying optimal stimuli is not straightforward. Walker et al. devised ‘inception loops’, which combine in vivo recordings of neuronal activity with in silico modeling. The researchers performed calcium imaging in layer 2/3 of the mouse primary visual cortex while showing the mice images of natural scenes. They then trained a deep convolutional network to predict neuronal responses from the images shown. Using the trained network, the researchers could identify images that excited individual neurons best. The researchers call these stimuli the most exciting inputs (MEIs). When presented to the mice, these MEIs indeed elicited strong responses in the neurons they were designed for. Inception loops may be particularly useful for studying neural computation in higher order brain areas, where optimal stimuli are even more elusive.