Markov Chain Monte Carlo Methods for Decoding Neural Spike Trains

Stimulus reconstruction or decoding methods provide an important tool for understanding how sensory and motor information is represented in neural activity. We address Bayesian decoding methods based on an encoding generalized linear model (GLM) [1, 2] that accurately describes how stimuli are transformed into the spike trains of a group of neurons. The log-concave GLM likelihood is combined with a prior distribution to yield the posterior distribution over the stimuli that possibly generated an observed set of spike responses. This posterior is log-concave so long as the prior is, meaning that the maximum a posteriori (MAP) stimulus estimte can be obtained using highly efficient optimization algorithms [3]. Unfortunately, however, the MAP estimate can have a relatively large average error when the posterior is highly non-Gaussian.