Figure 8From: Extracting novel information from neuroimaging data using neural fieldsSummary of the approach. We assume the measured signal is a mixture of predicted spectra, channel noise (a mixture of white and pink noise with weights alpha and beta) and Gaussian observation noise epsilon. Bayesian inference uses a Laplace approximation to the posterior density over unknown model parameters to obtain approximate conditional density and log-evidence – that are then used for inference on parameters and models respectively.Back to article page