Brain encoding enable brain-machine interface technologies

We work on the development of families of image-computable encoding models. These models help us decipher the computational role of various brain regions. Conversely, any accurately predicted brain representations can subserve various form of brain-reading technologies (decoding), whether in the visual domain (image decoding) or semantic domain (text and meaning decoding).

Multi-scale predictive brain representations uncover the common cause of brain activity

We aim to leverage multiscale brain recording (iEEG, EEG, fMRI) to enforce self-consistency in shared image-computable feature extractor backbone to better disentangle the role of the representational and measurement models in encoding models. This will also further our understanding of the link between measurement timescales (milliseconds in iEEG to seconds in fMRI).

Mental imagery as a keystone for understanding the relation between sensory and internally driven mental experiences

Visual mental imagery is the evocation of mental images in absence of a corresponding stimulus. We elaborate theories of mental imagery based on the principle of re-instanciation of evoked activity in a shared structure (generative model of visual recognition). Such models depend on and test the constraints imposed by the visual system structure.