On the Generation of Desired Outputs for Spike Neural Networks (SNN)
Tipo
Artigo de evento
Data de publicação
2023
Periódico
Lecture Notes in Networks and Systems
Citações (Scopus)
0
Autores
Menescal D.D.
de Castro L.N.
de Castro L.N.
Orientador
Título da Revista
ISSN da Revista
Título de Volume
Membros da banca
Programa
Resumo
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.In supervised learning algorithms, it is necessary to define an error function for the parameter adjustment process to take place. This function generation requires the input feature vectors and their respective desired outputs. In the context of neural networks, the network outputs are compared with the desired outputs so as to compute the error function. For standard networks, such as Perceptron, Adaline and others, the desired outputs are basically a class label or output value that will be directly used to calculate the network error. In the case of bioinspired networks, such as those using Leaky Integrate-and-Fire (LIF) neurons, their output are electrical impulses (spikes). In such cases, the electrical impulse has a built-in temporal dependence that does not occur for Perceptron neurons, thus representing a challenge to calculate the desired output values (spikes) for Spike Neural Networks (SNN). The purpose of this paper is to define an analytical solution to build the desired spikes for each category in a classification problem for a SNN. The computational challenge encountered to represent the dynamics of spike generation in bioinspired neurons will also be discussed, which has a direct impact on the objective of the proposed solution.