In the first part of the talk I will discuss the issue of autocorrelation in FMRI time series
("smoothness in the noise"), and describe one approach to modelling this in a flexible,
robust way, separately at each voxel. In the second part I will describe recent advances
in the modelling of the haemodynamic response function, using constrained optimised
basis functions iaBf, allowing more efficient temporal fitting of the FMRI signal.