Section: New Results

From the mesoscopic to the behavior scale

Participants: Axel Hutt, Laurent Bougrain, Eric Nichols, Maxime Rio, Carolina Saavedra, Louis Korczkovsky, Alexandre Martin, Pierre-Jean Morieux.

To link neural population activity to behavior, it is necessary to understand well the dynamic properties of population models which we have studied in general models [5] , [14] , [15] , [16] , [24] . To this end, we have analyzed a neural mass model [4] describing the neural population activity subject to synaptic anaesthetic action to explain characteristic signal features in measured EEG. The model explains the gain of power in the $\alpha -$ and $\delta -$frequency observed experimentally by a dynamic oscillatory instability (Hopf instability). The model considers a cortical population only and hence the result indicates that the experimental feature observed may originate in the cortex.

An extended population model considers not only the cortex but a feedback-loop to the thalamus. This model involves a delayed interaction. At first, we have studied the dynamics of delayed dynamical systems subject to additive stimuli [23] , [7] , [6] to learn more about the expected activity. Our first study of a linear thalamo-cortical feedback model [21] , [20] reveals the descriptive power of neural mass models to describe EEG under anaesthesia.

In order to learn more about the effect of anaesthetics on neural populations, we have participated in the data analysis of an experimental study on anaesthesia in animals [9] . Moreover we have started developing new data analysis techniques to extract novel features from EEG. In his doctoral thesis, Maxime Rio has developed a new method to detect transient amplitude synchronization in multi-variate time series in a subset of time series [1] . Carolina Saavedra has conducted wavelet analysis in her thesis to improve the denoising in BCI-relevant measured signals [2] . Another study [3] proposes a new recurrence plot-technique based on symbolic dynamics. It extracts spatio-temporally recurrence patterns in a multivariate dataset which reflect underlying neural recurrent dynamics.

The event-related potentials (ERP) in EEG are important markers of cognitive processes in the brain and serve as features to control interfaces in BCI. We have performed advanced studies to improve the detection of ERP [13] , [12] .