Théorique, numérique
Modern electrophysiological recording techniques have made it possible to simultaneously measure the activity of hundreds of neurons in the brain. To make sense of these massive recordings, appropriate techniques of data analysis and machine learning are necessary to reveal the relevant information contained therein. Because one of the main challenges of this objective are strong correlations between many interacting units, statistical mechanics is an excellent tool to tackle it, and has in fact been used in the realm of neuroscience for about four decades. In our work, we use maximum-entropy models, and extensions thereof to quantify the information about the stimulus contained in neural activity. This requires the fitting of models faithfully reproducing the statistics of numerous neurons, which is a hard task. However, maximum entropy models are static by nature, and this contrasts with the fundamentally dynamic nature of the stimuli in the real world. The topic of the internship project is therefore to extend the existing models to capture this temporal variability. Under our guidance, the intern will combine concepts stemming from statistical mechanics, information theory and machine learning to understand the data from ex-vivo retinas collected in our team.