The Processing and Storage of Information in Neuronal Memory Systems Across Time Scales

Humans and animals are able to store and recall information about past experiences across a variety of time scales. This process is accomplished by memory, which is implemented in the neuronal systems of the brain. These neuronal systems consist of a large number of electrically excitable cells, called neurons, which interact via contact points called synapses. The transmission efficacies of these synapses can be adapted by processes summarized as synaptic plasticity. Working memory (WM) describes the ability to store and to process information on time scales from seconds up to a minute and is important in many cognitive processes. The neuronal mechanisms underlying WM are still not understood. Some experimental and theoretical studies suggest that the neuronal system which implements WM stores information in the form of persistent activity of specific groups of neurons. These stable activity configurations are called attractor states. Other studies suggest that the information is stored in the form of complex temporal sequences of various activity patterns, so called transient trajectories. In this thesis, we show that the neuronal system implementing WM actually depends on both transient neuronal activity as well as distinct attractor states. Furthermore, we demonstrate that these attractor states may emerge in a self-organized way in the neuronal system implementing long-term memory (LTM) that stores information on time scales from hours to years. Finally, we develop a mechanism that may allow transient neuronal activity in the WM system to control long-lasting time-dependent output signals. First, we show that, different from human subjects, a model of a neuronal system which solely operates on transient activity dynamics is not able to solve a typical WM task with unpredictable temporal structure. Remarkably, the performance of this system is restored by introducing distinct attractor states into the system dynamics. Still, the transient trajectories in between these attractor states are required to enable non-linear time-dependent processing. Thus, the neuronal system which implements WM requires both transient dynamics and distinct attractor states. Second, we demonstrate that these attractor states can be created by groups of strongly interconnected neurons, so called cell assemblies (CAs), formed in the neuronal system implementing LTM. We show that CAs may be reliably formed and allocated to different stimuli by an interplay of two synaptic plasticity processes. Hence, the attractor states required by the WM system may emerge in a self-organized way in the LTM system. Third, we present a mechanism which enables a short transient signal to adapt the autonomously produced periodic output signal of neuronal systems called central pattern generators (CPGs). This mechanism allows to fast and precisely adapt the frequency of general oscillatory systems in a self-organized way based on the frequency of a short periodic stimulation. Thus, it enables short-lasting transient trajectories in the WM system to evoke long-lasting time-dependent neuronal signals. In summary, we show that to allow for WM that is robust with respect to unpredictable temporal structure and can perform complex non-linear processing, the underlying neuronal system has to rely on a combination of transient trajectories and distinct attractor states. These attractor states may emerge in a self-organized way in the LTM system and in CPGs