ERMITES_2012_M_Van_Gerven : « Percept Decoding with Sparse Latent Variable Models »

Functional MRI has an exquisite spatial resolution which offers a unique window into the brain . Using sophisticated machine learning techniques, researchers have shown that visual percepts can be decoded from fMRI BOLD responses. In this talk I will give an overview of the status of this field. Particularly, I will discuss the decoding problem from a probabilistic point of view and make the distinction between discriminative and generative approaches. Next to the discussion of models used by other researchers, I will present a number of approaches we have been developing within my group. These approaches range from simple Gaussian models to more complex deep architectures. Slides available online: http://sis.univ-tln.fr/~glotin/ERMITE...