Abstract: The retina represents the first stage of visual processing in many animal visual systems, including humans and mice. Retinal cells process light stimuli and fire electrical signals deeper into the brain for further processing. Retinal ganglion cells (RGCs) are the main site of visual processing in the retina (Masland, 2005). Three categories of RGCs have been established in mice, based on their function - on-cells respond only to ON temporal signals (transition from dark to light), off-cells respond only to OFF temporal signals (light to dark), and on-off-cells which respond to both. Recent research has found that there are numerous subcategories of these three main categories (Sanes & Masland, 2015). Additionally, UMD researchers have developed a novel mathematical model of RGC responses that incorporates nonlinear inputs (NIM) instead of the linear approximations that have been previously used (Cui et al, 2013). This document will show that unsupervised clustering of on-cell and off-cell temporal filters of nonlinear models finds subcategories of RGCs similar to previous classifications. This result shows that the NIM can find the same distinct subcategories as previous research. Thus the NIM is finding a similar set of solutions as other models, but the nonlinearity indicates that the NIM solutions are in a more general subspace. This finding shows the NIM to be more powerful than traditional linear models because of the implied generality. Future work includes clustering on-off-cell models and finding molecular or physiological bases for the distinct subcategories.
Voice: A formal voice will be used for the introduction/review of previous literature, but an informal first person voice will be used when discussing methods and results.
Reader response: This document simply shows the distinct subtypes of RGCs which have already been classified several times. It does not show how the NIM is "more powerful" or how these classifications contribute to the neuronal classification problem.
Voice: A formal voice will be used for the introduction/review of previous literature, but an informal first person voice will be used when discussing methods and results.
Citation style: Formal throughout.
Reader's profile: computational neuroscientists, computer/data scientists, neurophysiologists.
Reader response: This document simply shows the distinct subtypes of RGCs which have already been classified several times. It does not show how the NIM is "more powerful" or how these classifications contribute to the neuronal classification problem.