Exact Inference in Discontinuous Firing Rate Estimation Using Belief Propagation

概要

論文の詳細を見る
We apply image restoration methods to estimate the firing rate of neurons that include discontinuity according to the input stimulus. Image restoration methods are effective in estimating firing rate because images and firing rates can be respectively modeled by two-dimensional and one-dimensional Markov random field (MRF) models. Our method uses the line process, which was developed to detect edges in images, and to estimate discontinuous firing rates and entirely-unknown stimulus timings. We construct the firing rate estimation algorithm for our model using belief propagation (BP). BP gives the exact inference in the firing rate estimation because there is no loop in the first-order MRF. BP simultaneously estimates the discontinuous firing rate, unknown stimulus timings, and calculates a marginal likelihood value that enables us to estimate the hyperparameter based on an empirical Bayes method. By applying our method to synthetic spike trains, whose firing rates include discontinuity, we show that our method can be used to estimate firing rates and stimulus timings.
2009-06-15