Dynamic states of neural populations during cognitive computations

Neural responses and behavior are influenced by internal brain states, such as arousal or task context. Ongoing variations of these internal states affect global patterns of neural activity, giving rise to apparent variability of neural responses under the same experimental conditions. Uncovering dynamics of internal states from data proved difficult with traditional techniques based on trial averaged responses of single neurons. In this talk, I will describe our work leveraging multi-electrode neural activity recordings and computational models to reveal internal-state dynamics of neural populations during perception and goal-directed behavior. I will show how endogenous fluctuations of ensemble neural activity in the primate visual cortex depend on the global arousal and selective attention. The spatiotemporal structure of these fluctuations accounts for correlated variability across cortical layers and columns. I will then present a non-parametric framework for discovering neural population dynamics directly from the data without a priori model assumptions. Taking advantage of this flexible, yet intrinsically interpretable framework, I will demonstrate a distinction between good data prediction and accurate interpretation of the model, and propose a strategy for deriving models with correct interpretation.