Export Reviews, Discussions, Author Feedback and Meta-Reviews

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. In the last few years, there has been considerable interest in the neuroscience community in decoding high-level latent brain states from noisy neural recordings. One such state variable is attention: primates and other high-level organisms can preferentially distribute resources to encode and process a selective set of incoming stimuli in a way that is typically not externally visible. In the case of auditory selective attention, empirical studies have identified a set of neural variables that can be measured that provide information about which sound a subject is currently directing attention to. Some of these variables can be measured via magnetoencephalography (MEG). This study capitalises on these prior observations to build a statistically-principled decoder of human auditory attentional states under competing-speaker auditory stimulation. The authors explain the underlying measurements, represent the problem as one of Bayesian inference, then present an EM-based inference procedure for inferring the latent variables.