Bayesian Analysis of Neural Firing Rates

• Activity Number: 465
• Type: Topic Contributed
• Date/Time: Thursday, August 7, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Stat. Sciences
• Author(s): Anne Smith*+ and Sylvia Wirth and Marianna Yanike and Wendy Suzuki and Emery Brown
• Companies: Massachusetts General Hospital and New York University and New York University and Center for Neural Science and Harvard University
• Keywords: state-space model ; latent process model ; neural encoding

Abstract

State-space or latent process models have been used extensively in many fields of science and engineering. We describe application of Bayesian hierarchical state-space modeling to the analysis of point process data acquired in neurophysiological experiments. The analysis is motivated by stimulus-response experiments where the experimenter observes cell firing during the application of an external (implicit) stimulus over many repeated trials. Our model comprises an observation equation describing how the latent state is observed and a state equation describing the changes in state over time. A hierarchy is imposed to permit a distribution of dependencies between each trial and the latent state. We apply our analysis to single neuron recordings from the monkey hippocampus during a location-scene association task. Agreement between the model and neural point process data is evaluated using a Kolmogorov-Smirnov test based on the time rescaling theorem.