Incorporating neural response variability into models for neural coding

Abstract

One of the primary challenges facing neuroscientists is understanding how information is represented in neural circuits. These representations provide insight into the computations performed by individual neurons and neural circuits. Complicating this endeavor is the variability in neural responses: repeatedly presenting the same stimulus does not elicit identical responses. Although variability is often treated as a nuisance that obscures relevant features of the neural response, the origin and nature of this variability have meaningful implications for how we understand computations in neural circuits, as well as the perceptions and behaviors that rely on these computations. Here, I present multiple approaches to characterizing the role that variability plays in how information is processed in the nervous system. I first examine a widely used class of models, generalized linear models, and evaluate its ability to capture response features observed in biological neurons. I then examine how variability in the responses of retinal ganglion cells changes under different conditions and propose a new model that accounts for responses under both conditions. Finally, I examine how the origin of noise in neural circuits influences optimal coding strategies.