Boolean Neural Networks
Neural Networks, which are the foundation of every human brain, are very peculiar structures. Their functioning can produce amazing results. Learning about them and the way they work, however, is an uphill task.
To respond to this challenge, several models that represent the functionality of the neural networks were created, for instance, [6] and [14] . These models are successfully used in niche applications. However, performance of these models is lower than of a human brain in comparable tasks. Besides that, energy consumption is larger by several factors.
As a part of the thesis a novel neuron model was developed. It is motivated by the goal of extension of the current use of artificial neural networks that represent the ideas of how biological networks work. A model that works precisely and fast can be used in technical solutions that serve people. The novel neuron model is based on principles of the neuron model described in [11] and [18]. It is extended with data structures as well as processes that optimize functioning of the network.
The novel model has dendritic trees in place which are prototypes of a biological version of these trees. In addition, it has a centralized control for Boolean values in that tree and employs Boolean functions. In order to analyze the model, it was implemented as a software simulator of neural activity. It is written in C++ which leads to swift execution.
Moreover, it has a great potential to yet improve the performance by being ported and executed on a special programmable chip, such as a field-programmable gate array (FPGA).
Master’s Thesis in Informatics:
Design and Analysis of a Novel Boolean Neuron Model
Sergey Nasonov
Department of Computer Science, Technical University of Munich
