Visual Computation

Our interest in using the fly visual system to model computation stems from the fact that flies are capable of some of the same fundamental visual computations performed in mammalian brains.  Moreover, the statistics of the visual environment have not changed over evolutionary time, meaning that visual systems that perform common tasks, such as motion perception, are constrained by the same basic constraints.  Indeed, our recent work has demonstrated functional parallels between the fly and vertebrate visual systems at multiple levels.  Given this, the numerically smaller, anatomically more stereotyped brains of invertebrates afford significant advantages for exploring small, specific circuits of defined connectivity that implement critical algorithmic steps.

We focus on the neural circuits that process two very different visual cues, namely motion and polarized light.  These two classes of stimuli were chosen for three reasons.  First, as a minimal form of spatiotemporal correlation, motion vision has long been considered a paradigmatic computation, representative of many brain functions.  Second, polarized light detection is thought to rely on neural circuts that implement opponency, a computation of broad interest.  Finally, these two stimulus types enter the brain through distinct photoreceptors but affect the same behavioral output, locomotion.  By presenting both cues in conjunction, we will be able to investigate the neural mechanisms underpinning a long-standing mystery, namely how different types of cues are bound together in a single percept.

To understand these questions, we take advantage of two broad types of approach, one that uses behavior as a read-out of the integrated performance of the entire circuit, and one that uses imaging based approaches to monitor the activities of specific neuron types directly.  Previous work in the lab showed that, like humans, motion processing in flies is conducted in two at least partially separate pathways, one specialized for moving light edges, and one specialized for moving dark edges.  We have then gone on to identify and characterize neural components in both pathways, and have uncovered many of the key computational steps in motion processing.  

Previous work in the lab has also defined the peripheral neural components involved in the detection of polarized light, work that sets the stage for studies that look at the interaction between both visual cues.

Current work in the lab is focused on defining exactly the algorithmic basis for motion processing, determining how the key computational steps are implemented at the cell biological and molecular level, and unraveling how the circuits that process visual information evolve and are shaped by different environmental conditions.