Brain decoding and inverse inference in fMRI.
A new approach for brain decoding, called inverse inference (or brain-reading), has been introduced recently [Dehaene 98, Cox 03]. This method relies on statistical learning tools, and more precisely on pattern recognition approaches. The main idea is to consider the fMRI analysis as a pattern recognition problem, i.e. using a pattern of voxels to predict a behavioral, perceptual or cognitive variable. In this approach, the accuracy of the prediction can be used to validate (or invalidate) that the pattern of voxels used in the predictive model is implied in the neural coding. In short, inverse inference is an approach for decoding the neural coding.
-
Talks on Brain decoding and inverse inference in fMRI.
- Selected talks on Brain decoding and inverse inference in fMRI.
-
Brain decoding and inverse inference in fMRI: A brief introduction
- The overall aim of this research project is the development of statistical learning methods that take into account the characteristics of fMRI data, and that can be used for inverse inference. From an experimental point of view, we particularly focus on the understanding of the human visual cortex, but the presented frameworks can be used to study any brain system.
-
Articles on Brain decoding and inverse inference in fMRI.
- Articles on Brain decoding and inverse inference in fMRI.
-
Multi-Class Sparse Bayesian Regression for brain decoding
- We briefly detail here a novel Bayesian regularization approach, called Multi-Class Sparse Bayesian Regression (MCBR), based on a multi-class organization of the features, with the aim of adapting the amount of regularization to the informative content of each feature class.
-
Gallery
-
Supervised clustering for fMRI inverse inference
- We briefly detail here an original contribution, called supervised clustering, for feature agglomeration in fMRI inverse inference, that handles both the spatial structure of the images, and the multivariate nature of the signal.
-
Total variation regularization
- We propose to use the 1 norm of the image gradient, a.k.a. its Total Variation (or TV ), as regularization.

