Sinusoidal Neural Networks: Towards ANN that Learns Faster

Tekin Evrim Ozmermer


If everything is a signal and combination of signals, everything can be represented with Fourier representations. Then, is it possible to represent a signal with a conditional dependency to input data? This research is devoted to the development of Sinusoidal Neural Networks (SNNs). The motivation to develop SNNs is to design an artificial neural network (ANN) algorithm that can learn faster. A short review of the history of biological neurons helps to identify components that should be redesigned in ANNs. After the components are identified, a new neural network algorithm called SNN is proposed. Experiments are conducted to show the practical results of the algorithm. According to the experiments, the proposed neural network can reach high accuracy rates faster than the standard neural networks, while an interesting generalization capacity is obtained for the developed algorithm. Even though the promising results are achieved, further research is necessary to test if SNNs are capable of learning faster than existing algorithms in real-life cases.


Artificial Neural Networks; Fourier Neural Networks; Periodic Functions; Activation Function; Node Operation

Full Text:


DOI: 10.7250/csimq.2020-23.04


  • There are currently no refbacks.

Copyright (c) 2020 Complex Systems Informatics and Modeling Quarterly