Large-scale neuronal networks: from simulation technology to the study of plasticity-driven cognitive processes

The brain is one of the most intricate systems and the quest for simulating the neuronal dynamics has led to the development of several approaches and simulation tools able to represent the behavior of portions of the brain with different levels of detail. Among the different techniques that we can adopt to shed light on neuronal dynamics, mean-field models and spiking neural networks are two of the most relevant, and are introduced in Part I of this doctoral thesis. In particular, spiking neural network models have become an effective tool to study brain functions as they capture several aspects of natural neural networks, with every neuron being characterized by a membrane potential and a mechanism to emit electrical pulses for communication with other neurons. While mean-field approaches are suitable for the simulations of models of the entire brain with population resolution, spiking neural networks can simulate portions of the brain at cellular level. However, recent computing technologies are paving the way for large-scale simulations through the usage of cutting-edge supercomputer clusters, and it is of fundamental importance for computational neuroscientists to have tools able to take advantage of these technologies. In recent years, Graphical Processing Units (GPUs) established themselves as promising hardware to be employed for such simulations, thanks to their high degree of parallelism, and several GPU-based simulation codes have been developed. In Part II of this thesis, we describe the GPU code for spiking neural network simulations NEST GPU, which can efficiently exploit GPU hardware spanning from consumer GPUs to data-center cards employed in MPI-GPU clusters. The thesis is devoted both to evaluate the performance of such a simulator in the simulation of neuroscientifically relevant models, and, most importantly, to validate the results of the neuronal dynamics with respect to established spiking network simulators such as NEST. To better understand the link between brain functioning and high-level cognitive processes with low-level neuronal activity, there is the need to provide realistic models both for the neurons and the synapses. Indeed, there is broad consensus in the neuroscientific community that synaptic mechanisms, such as short-term synaptic plasticity and structural synaptic plasticity underlie cognitive processes like working memory and learning. Part III of this thesis is devoted to developing simulation and theoretical frameworks that shed light on the possible relation between these synaptic mechanisms and the previously mentioned cognitive processes. In particular, we focus on the simulation of a working memory spiking network driven by short-term plasticity, which is believed to be responsible for the activity-silent mechanisms that characterize working memory networks, and we also present a theoretical framework able to describe a learning process mediated by structural synaptic plasticity, evaluating the memory capacity of the network as a function of the simulation parameters. This thesis aims to start facing the challenge of studying high-level cognitive processes through simulations of large-scale neuronal networks. In a framework in which computing technologies are opening to the realm of large-scale simulations through the usage of GPU clusters, there is a need for simulators capable of exploiting this fast-growing hardware being efficient and reliable. Additionally, modeling neuron and synaptic scale mechanisms can shed light on their impact on high-level cognitive processes such as learning and memory and, together with large-scale simulations at neuron resolution, it would be possible to estimate the relation of these mechanisms and the dynamics of neuronal networks representing a significant portion of the brain. These works are oriented toward the development of more detailed network models, which will pave the usage of these tools in medicine as support for novel therapies.