Une des capacités cérébrales les plus fondamentales, qui est essentiel pour tous les fonctions cognitifs de haut niveau, est de garder des informations pertinentes de tâche pendant les périodes courtes de temps; on connaît cette capacité comme la mémoire de travail (WM). Dans des décennies récentes, accumule là l'évidence d'activité pertinente de tâche dans le cortex préfrontal (PFC) de primates pendant les périodes de "delay" de tâches de "delay-response", impliquant ainsi que PFC peut maintenir des informations sensorielles et ainsi la fonction comme un module de WM. Pour la récupération d'informationssensorielles de l'activité de réseau après que le stimulus sensoriel n'est plus présent il est impératif que l'état du réseau au moment de la récupération soit corrélé avec son état au moment de la compensation de stimulus. Un extrême, en vue dans les modèles informatiques de WM, est la coexistence d'attracteurs multiples. Dans cette approche la dynamique de réseau a une multitude d'états stables possibles, qui correspondent aux états différents de mémoire et un stimulus peut forcer le réseau à changer à un tel état stable. Autrement, même en absence d'attracteurs multiples, si la dynamique du réseau estchaotique alors les informations sur des événements passés peuvent être extraites de l'état du réseau, à condition que la durée typique de l'autocorrélation (AC) de dynamique neuronale soit assez grande. Dans la première partie de cette thèse, j'étudie un modèle à base d'attracteur de mémoire d'un emplacement spatial, pour examiner le rôle des non-linéarités de courbes de f-I neuronales dans des mécanismes de WM. Je fournis une théorie analytique et des résultats de simulations montrant que ces nonlinéarités, plutôt que les constants de temps synaptic ou neuronal, peuvent être la base de mécanismes de réseau WM. Dans la deuxième partie j'explore des facteurs contrôlant la durée d'ACs neuronales dans ungrand réseau "balanced" affichant la dynamique chaotique. Je développe une théorie de moyen champ (MF) décrivant l'ACs en termes de plusieurs paramètres d'ordre. Alors, je montre qu'en dehors de la proximité au point de transition-à-chaos, qui peut augmenter la largeur de la courbe d'AC, l'existence de motifs de connectivité peut causer des corrélations de longue durée dans l'état du réseau. / One of the most fundamental brain capabilities, that is vital for any high level cognitive function, is to store task-relevant information for short periods of time; this capability is known as working memory (WM). In recent decades there is accumulating evidence of taskrelevant activity in the prefrontal cortex (PFC) of primates during delay periods of delayedresponse tasks, thus implying that PFC is able to maintain sensory information and so function as a WM module. For retrieval of sensory information from network activity after the sensory stimulus is no longer present it is imperative that the state of the network at the time of retrieval be correlated with its state at the time of stimulus offset. One extreme, prominent in computational models of WM, is the co-existence of multiple attractors. In this approach the network dynamics has a multitude of possible steady states, which correspond to different memory states, and a stimulus can force the network to shift to one such steady state. Alternatively, even in the absence of multiple attractors, if the dynamics of the network is chaotic then information about past events can be extracted from the state of the network, provided that the typical time scale of the autocorrelation (AC) of neuronal dynamics is large enough. In the first part of this thesis I study an attractor-based model of memory of a spatial location to investigate the role of non-linearities of neuronal f-I curves in WM mechanisms. I provide an analytic theory and simulation results showing that these nonlinearities, rather than synaptic or neuronal time constants, can be the basis of WM network mechanisms. In the second part I explore factors controlling the time scale of neuronal ACs in a large balanced network displaying chaotic dynamics. I develop a mean-field (MF) theory describing the ACs in terms of several order parameters. Then, I show that apart from the proximity to the transition-to-chaos point, which can increase the width of the AC curve, the existence of connectivity motifs can cause long-time correlations in the state of the network.

Pas de résumé en français / It is generally maintained that one of cortex’ functions is the storage of a large number of memories. In this picture, the physical substrate of memories is thought to be realised in pattern and strengths of synaptic connections among cortical neurons. Memory recall is associated with neuronal activity that is shaped by this connectivity. In this framework, active memories are represented by attractors in the space of neural activity. Electrical activity in cortical neurones in vivo exhibits prominent temporal irregularity. A standard way to account for this phenomenon is to postulate that recurrent synaptic excitation and inhibition as well as external inputs are balanced. In the common view, however, these balanced networks do not easily support the coexistence of multiple attractors. This is problematic in view of memory function. Recently, theoretical studies showed that balanced networks with synapses that exhibit short-term plasticity (STP) are able to maintain multiple stable states. In order to investigate whether experimentally obtained synaptic parameters are consistent with model predictions, we developed a new methodology that is capable to quantify both response variability and STP at the same synapse in an integrated and statistically-principled way. This approach yields higher parameter precision than standard procedures and allows for the use of more efficient stimulation protocols. However, the findings with respect to STP parameters do not allow to make conclusive statements about the validity of synaptic theories of balanced working memory. In the second part of this thesis an alternative theory of cortical memory storage is developed. The theory is based on the assumptions that memories are stored in attractor networks, and that memories are not represented by network states differing in their average activity levels, but by micro-states sharing the same global statistics. Different memories differ with respect to their spatial distributions of firing rates. From this the main result is derived: the balanced state is a necessary condition for extensive memory storage. Furthermore, we analytically calculate memory storage capacities of rate neurone networks. Remarkably, it can be shown that crucial properties of neuronal activity and physiology that are consistent with experimental observations are directly predicted by the theory if optimal memory storage capacity is required.