Application of machine learning to fMRI/EEG data?

Machine learning has been applied to neuroimaging data since the early 90's and has become an explosive field.

The specific data that you refer to is resting state data (i.e. no specific task) and there has been much less machine learning applied to that type of data. The majority of machine learning applications to fMRI are "mind reading" problems, that are solved with supervised learning. In this case you have a label describing the subjects behavior at the time that the fMRI data was acquired, and you goal is to learn a classifier capable of decoding the measure of behavior from the brain data. For resting state data, the subject is just asked to lie there quietly and not think of anything. Thus, there are no labels for supervised learning. Supervised learning from resting state data can be used for "disease state prediction" or "phenotype prediction" in which your labels are assigned to the subject, and the fMRI data is somehow summarized into an appropriate feature set per subject. Examples of this in literature are Disease state prediction from resting state functional connectivity and Prediction of Individual Brain Maturity Using fMRI. There was a competition in 2011 to learn the best biomarker to prediction ADHD diagnosis, and that data is still available online: The ADHD-200 Sample. If you are not savvy with functional neuroimaging you can download preprocessed and feature extracted versions of this data from: The ADHD-200 Sample - Introduction.

Additionally researchers use resting state data to measure "functional connectivity" which presumes that brain regions which have correlated time series are meaningfully connected. Functional connectivity is typically measured using bivariate measures of statistical dependency such as pearson's correlation, clustering algorithms, or ICA. There have been a few researchers (Gael Varoquaux being one of them) that employ regularized regression and predictive modeling techniques to estimating functional connectivity. For example of this work, see Measuring the Consistency of Global Functional Connectivity Using Kernel Regression Methods or Brain covariance selection: better individual functional connectivity models using population prior.

Have a look at www.pymvpa.org -- quite a few tutorial topics on application of machine learning in fMRI. Publications on the site would point you to relevant (fresh and not so) articles. Also in http://www.frontiersin.org/neuroinformatics/10.3389/neuro.11.003.2009/abstract we have applied PyMVPA machinery across EEG/MEG/fMRI/extra-cellular. And if you are interested in a joint fMRI/EEG analysis (I called it transfusion) -- checkout my PhD thesis (section 4.4, never managed to finilize that publication draft to submit)

Many scikit-learn developers are applying machine learning to brain mining on a day to day basis. For instance you can have a look at this tutorial on fMRI inverse inference.

Gael Varoquaux also recently presented the following talk on Python for brain mining: (neuro)science with state of the art machine learning and data visualization during the SciPy 2011 conference.

Edit: here is a clickable link to the video of Gael's talk.

This is the only paper on fMRI stuff I've seen in the last few years (though that's not to say it isn't a busy field): Generative versus Discriminative Training of RBM's for classification of fMRI images.