Evan Schaffer

Wednesday 27th of June 2018

Time:4.00pm

Ground Floor Seminar Room

25 Howland Street, London, W1T 4JG

Learning and generalizing with random sensory representations

The ability to generalize is vital to performance in any task because natural variability implies that the exact same stimulus is never experienced twice. Moreover, brain regions involved in a behavior must generalize in a similar manner. This presents a problem in olfaction, because neurons in the mouse piriform cortex receive input from an apparently random collection of olfactory glomeruli, implying that the representation of odors will differ across brain hemispheres. How is consistency of olfactory generalization possible between different randomly-wired cortices? In the first half of my talk, I will describe a model in which we demonstrate that multiple readout units, each connected to an independent piriform, produce highly correlated output to any novel odor after training on a single stimulus. We next modeled binary choices by imposing a threshold on each readout and examined the choice agreement across a large number of odors. Whereas odor discrimination and categorization require far fewer neurons than reside in piriform, we find that the ability of the model to support consistent choices across brain hemispheres requires the full complement of piriform neurons. To test predictions of our model that are not readily testable in the mouse, we exploit experimental data from the Drosophila mushroom body. We find that our model can predict the degree of correlation in vivo between pairs of mushroom body output neurons with striking accuracy. Following the observation that data from the fly offers unparalleled opportunities to test predictions from network models, in the last part of my talk, I will describe our recent work performing whole-brain imaging in adult behaving Drosophila. Using SCAPE, a variant of light-sheet microscopy, we can image the entire fly brain with single-cell resolution at rates exceeding 10 brain volumes per second. This allows us to examine neural dynamics distributed throughout the brain of an animal engaged in complex behavior. I will describe our ongoing work imaging flies transitioning into and out of behavioral states such as simple locomotion and internal states such as fear. Lastly, I will describe our efforts to image flies engaged in an olfactory associative learning task.