• spiking neural network models
• power spectral analysis
• neural synchronization
• cortical oscillations

More realistic pictures of the structures and functions of the nervous systems are now possible, due to the notable progresses in experimental/computational techniques and tools.
Advancements of knowledge take place also for the more complex brains, such as the mammals ones.
A systematic catalogue of neurons and their connections has been pursued and is still in rapid development.
Thereby, more detailed models of large parts of the brain have become a viable endeavor.
The visual cortex, since the seminal works of Hubel and Wiesel, has always been a key benchmark for understanding how the brain works.
Moreover, since humans are mainly visual animals, it is also a way to understand how we behave.
In this thesis we have developed a novel spiking neurons model of layer 4, the most responsive layer of primary visual cortex.
We have taken track of the different neural types that have been discovered so far in visual cortex, but reducing their number to three main types.
In particular, with respect to previous models, which considered only a generic class of inhibitory neurons, we have introduced a more detailed description of two different kinds of inhibitory neurons (Parvalbumin and Somatotstatin), with specific features and distinct dynamics.
Their parameter values have been taken from the most updated database.
As first step, we have characterized the single neural model dynamics.
Then, we have built and characterized two distinct (Parvalbumin and Somatotstatin) inhibitory networks.
For Parvalbumin network, the addition of an excitatory neural class has been necessary, in order to obtain a clear γ rhythm in neural activity, according to experimental data.
For each model, we have studied the network responses following parameter variations, in order to refine the connectivity matrix.
Finally, the second inhibitory population is added to the inhibitory-excitatory network, determining the three-neural networks connectivity matrix.
The use of simplified single-neuron model and randomly interconnected network has allowed to perform analytical studies of the system, in addition to simulations.
The resulting full network is able to reproduce the most relevant features of visual cortex dynamics.
The power spectrum analysis of neural population activity shows a peak in the low γ (30-70) Hz band, confirming previous results even in this more realistic network.
Crucially, the model has reproduced the fact that distinct neural subtypes are responsible of diverse population oscillations.
In particular Somatostatin neurons display stronger low-frequency oscillations comparable to the β (12 − 30) Hz band, while Parvalbumin neurons have the main role in the γ band oscillations. These results pave the way to a clearer understanding of inhibitory mechanisms at play in healthy and pathological processing of the visual information.