PPG - Neurologia - Neurociências

URI Permanente para esta coleção

https://hdl.handle.net/11600/61157

Navegar

Navegando PPG - Neurologia - Neurociências por Orientador(es) "Almeida, Antonio Carlos Guimaraes De [UNIFESP]"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Acesso aberto (Open Access)
    Redes neuronais realísticas e aprendizado
    (Universidade Federal de São Paulo (UNIFESP), 2019-03-28) Depannemaecker, Damien Thomas [UNIFESP]; Almeida, Antonio Carlos Guimaraes De [UNIFESP]; http://lattes.cnpq.br/2815105916715425; http://lattes.cnpq.br/6928800226748514; Universidade Federal de São Paulo (UNIFESP)
    Objective: Learning in the neural network inspired by brain tissue has been studied for the application of machine learning. However, studies are mainly based on the concept of synaptic weights adaptation and other aspects of neuronal interaction as non-synaptic mechanisms are neglected. This study aims to evaluate the learning processes in the assemblies of neurons and bring new insights to the computational application, as well as contributing to the understanding of the functioning of biological neural tissues. Methods: Realistic computational models based on experimental observations available in the literature or performed in Experimental and Computational Neuroscience (LANEC) of UFSJ were used for descriptions of neuron assemblies. These models were compared to methods of artificial neural networks used for machine learning, and to phenomenological models the neural activities. Results: Realistic models have better processing capacity, converging more quickly than artificial neural networks to solve the basic tasks studied. Nonsynaptic parameters have been identified, influencing the learning capacity, which are electrodiffusion along with tortuosity. It has been shown that tissue morphology also has a significant influence. The implementation of non-synaptic effects in phenomenological models maintains these properties. Finally, we observed electrophysiological activity, showing a mixture of different electrophysiological patterns, and a high synchronism. Conclusion: Neural tissues are complex systems where numerous parameters interfere, among which the electrodiffusion together with the tortuosity contributing to the learning capacity. Non-synaptic mechanisms affect neuronal excitability favoring specific pathways to respond to certain stimuli. These effects may vary depending on tissue morphology, revealing a relationship between morphology and function. Phenomenological models may become more efficient if non-synaptic mechanisms are included, thus improving their application for machine learning. The measure of synchronism in both models shows that the higher efficiency appears with high synchronism and near the epileptiform crisis threshold.