The mouse has become a major model for studying vision, because of the genetic, imaging, and molecular tools available. However, a basic conundrum has arisen: mice are capable of sharp, visually-mediated behaviors, such as accurate prey capture, but when assessed using standard methods (e.g., tunings curves, reverse correlation) and standard stimuli (moving gratings with low spatial frequencies), they appear to have very poor vision. We set out on the task of finding a class of artificial stimuli that are capable of illuminating previously-unknown aspects of these cells. By exploring the domain of moving dot patterns consisting of single or multiple dots linearly-aligned so as to have clear direction and/or orientation, we created visual flow fields with a richer geometry than gratings, and at the same time more ecologically relevant.
More complex stimuli result in more complex responses from the recorded cells, thus requiring more careful analysis. We present new approaches for analyzing neuronal spiking data using traditional machine learning techniques. First, we investigate how the neural population activity varies when the animal is exposed to different stimuli by embedding the population spikes in diffusion coordinates. We also adopt an unsupervised approach that is capable of organizing cells' responses at the individual level into meaningful clusters based on their 2-D peristimulus time histograms. By using a novel stimulus, together with new ways of analyzing spiking activity, we pave the way for a more complete characterization of the functional role of cells in the visual cortex.