Structured neural variability as a substrate for adaptive behavior.
Abstract: Across brain regions and species, one key feature of neural activity is that responses are highly variable. Typically this variability is viewed as noise, which leads to a view in which a big, perhaps the biggest, challenge for brain computation is compensating for its own internal noise. In contrast, we argue there may be computational advantages to having a seemingly ‘noisy’ brain. In this talk I will discuss our recent theoretical results on how low-dimensional structured noise can be used to dynamically route task-specific information between neural populations. I’ll describe our efforts in confirming this idea 1) experimentally, using recordings of population activity in the visual stream of primates during blocked task changes and 2) computationally, by incorporating the same mechanism in machine learning models.