Revealing nonlinear neural decoding by analyzing choices


February 15, 2017 - 1:00pm
Northwest 243
About the Speaker
Xaq Pitkow
Speaker Affiliation: 
Baylor College of Medicine & Rice University

Abstract: Most natural task-relevant variables are encoded by neural populations in the early sensory cortex in a form that can only be decoded nonlinearly, largely due to task-irrelevant variation. Yet nonlinear population codes are rarely studied and poorly understood, despite being a core motif of the brain. Here we provide a theory of feedforward nonlinear population codes that obey fundamental limitations on information content that are inherited from the sensory periphery. We use this theory to provide a simple, practical formula using choice-related activity in small populations of neurons that tests if the brain uses its information optimally. If not, then we can identify essential properties of the suboptimal nonlinear computations, without needing to identify every detail of the computational strategy.