Mixed-effects inference for classification group studies
Multivariate classification algorithms are powerful tools for predicting cognitive or pathophysiological states from neuroimaging data. However, meaningful inference requires models that explicitly account for fixed-effects (within-subjects) and random-effects (across-subjects) variance components. While models of this sort are standard practice in mass-univariate analyses of fMRI data, they have not yet received much attention in the decoding literature.
This software closes this gap by providing full Bayesian mixed-effects inference for multivariate classification studies. It includes step-by-step instructions and is extremely easy to use. The software includes:
- a MATLAB toolbox which comprises both MCMC sampling algorithms and computationally more efficient variational Bayes (VB) approximations;
- an R package which focuses on variational Bayesian algorithms.
- K.H. Brodersen, J. Daunizeau, C. Mathys, J.R. Chumbley, J.M. Buhmann, & K.E. Stephan. Variational Bayesian mixed-effects inference for classification studies (2013). NeuroImage (in press). doi:10.1016/j.neuroimage.2013.03.008.
- K.H. Brodersen, C. Mathys, J.R. Chumbley, J. Daunizeau, C.S. Ong, J.M. Buhmann, & K.E. Stephan. Bayesian mixed-effects inference on classification performance in hierarchical datasets (2012). Journal of Machine Learning Research, 13, 3133-3176.
- K.H. Brodersen, C.S. Ong, J.M. Buhmann, & K.E. Stephan (2010). The balanced accuracy and its posterior distribution. ICPR, 3121-3124.