Soutenance de thèse: Jerôme EMONET (INRIA)

Devant le jury composé de :
Frédéric ALEXANDRE, Directeur de recherche, Université de Bordeaux
Viktor JIRSA, Directeur de recherche, Université d’Aix-Marseille
Michael BERRY, Professeur agrégé, Université de Princeton
Joana CABRAL, Professeur assistant, Université de Minho
Bruno CESSAC, INRIA Directeur de recherche, Université Côte d’Azur
Frédéric CHAVANE, Directeur de recherche, Université d’Aix-Marseille
Alain DESTEXHE, NeuroPsi Directeur de recherche, Université Paris-Saclay.

Résumé :
The visual system is able to process complex movements quickly and efficiently in rich visual scenes thanks to a processing taking place in the retina, the LGN, and the visual cortex. Especially, lateral connectivity enables the generation of waves that only appear when a moving object is present. In our work, we are particularly interested in the involvement of these waves in two mechanisms of motion vision: anticipation and saccadic omission. Anticipation is the name given to the set of mechanisms in the visual system to bring forward in time the representation of the stimulus in the retina or cortex. Their aim is to compensate for the 100ms delay in visual processing in the retina. With such a delay, we would not be able to catch a moving object. Saccadic omission is a mechanism for ignoring very rapid eye movements known as ocular saccades. These saccades enable us to explore the visual scene although we do not perceive them (saccadic omission). The current hypothesis is that saccadic omission only occurs in the presence of eye movements. However, an alternative hypothesis supports the indispensable role of perisaccadic stimuli. Experiments on saccades without eye movement have shown that in the presence of perisaccadic stimuli (with a sufficiently high frame rate, 1440 Hz) individuals perceive a slower, sharper moving bar, with reduced amplitude and no smear. This creates a movement that the visual system could easily erase. This process could be caused by the propagation of suppressive waves, already exhibited in apparent movements, through lateral connectivity in the cortex. In this potential mechanism of saccadic omission the most recent representation of the moving object inhibits older ones thanks to suppressive waves.

The aim of this thesis is to study the mechanisms of anticipation and saccadic omission from a modelling perspective To this end, we are using a retinal-cortical model known as the ‘Chimera model’. It first contains a retinal model made up of bipolar, amacrine and ganglion cells interconnected, with gain control on the bipolar and ganglion cells. The retinal output is sent to a cortical model, a network of cortical columns, each corresponding to mean-field equations, capable of producing the signal intensity of a pixel in voltage-dependent imaging. The output of the model can therefore be compared with biological experiments. This retino-cortical model has been implemented in the Macular plateform created at INRIA. This thesis introduces the Chimera model, review some of its dynamical properties, and discuss its implementation, before studying anticipation and saccadic omission. After a calibration of the model reproducing experiments made in F. Chavane lab, we show how different experimental (bar speed, contrast) and physiological (retinal output intensity, amacrine cells connections, gain control, cortical connections intensity, ...) impact anticipation at the level of the retina and in the cortex. For saccadic omission, we first reproduce the suppressive waves observed with apparent movement before studying the retino-cortical response of stimuli corresponding to movement, with or without perisaccadic stimuli, displayed at a refresh rate of 60Hz (classical video projectors) or 1440Hz (refresh speed at which a smear effect is perceived). We also study the impact of static phases and the potential role of suppressive wave propagation.

Mots-clés : Computational neuroscience, Motion vision, Retino-cortical model, Visual anticipation, Saccades