Bayesian reconstruction of perceptual experiences from human brain activity

A method for decoding the subjective contents of perceptual systems in the human brain would have broad practical utility for communication and as a brain-machine interface. Previous approaches to this problem in vision have used linear classifiers to solve specific problems, but these approaches were not general enough to solve complex problems such as reconstructing subjective perceptual states. We have developed a new approach to these problems based on quantitative encoding models that explicitly describe how visual stimuli are (nonlinearly) transformed into brain activity. We then invert these encoding models in order to decode activity evoked by novel images or movies, providing reconstructions with unprecedented fidelity. Here we briefly review these results and the potential uses of perceptual decoding devices.