Helen Kang Rose Hills
Estimating Integrated Information from Non-Gaussian Multivariate Time Series
Information theory has been key to our understanding of the brain, by elucidating mechanisms of neural coding in the brains sensory periphery. Recent mathematical work has derived a new measure called integrated information, which can quantify how much information emerges at the level of an entire network. Integrated information has already been used to study how the structure of brain networks might underpin large-scale information flow and may help explain mechanisms of information processing in both healthy and pathological brain states. Currently, integrated information can only be calculated for neural time-series data if those data are multivariate Gaussian. This limits the application of integrated information to most types of neural data, which are typically non-Gaussian. I will verify a method I have developed to estimate integrated information from non-Gaussian time series simulated from brain-like networks. I will also use this method to test the long-held but unverified hypothesis that large-scale cortical information integration should decrease during unconscious states, such as anesthesia, seizures, and slow-wave sleep.