Overview

We start in section 2 by describing how the GLM in a fully Bayesian framework allows us to introduce soft constraints on linear combinations within the GLM. We then discuss how we choose a sensible basis set, and determine the required basis set constraints. In section 3 we describe how we perform approximate Variational Bayesian inference on the model. In section 4 we use the model on artificial null data to demonstrate the effect of the HRF constraints. Then in section 5 we describe how we can use spatial mixture modelling to produce probabilities of activation which takes advantage of the extra sensitivity produced from the HRF constraints. Finally, in section 6 we demonstrate this increased sensitivity on an audio-visual dataset.