STDP-based Growing Neural Gas for Hierarchical Data Representation and Neural Networks Fusion

Abstract

Research in neurophysiology led to greater comprehension of the mechanisms involved in synaptic plasticity and a better understanding of the connections between neurons. That knowledge has been transposed to the field of ANNs (Artificial Neural Networks), leading to more incredible advancements in computational models, especially concerning the second generation of Neural Networks, which roughly coincides with “Deep Learning.”. Our research focuses on third-generation, spiking Neural Networks, and the motivation behind this work is creating neural networks able to communicate with each other. This work presents a novel, unsupervised, bio-inspired, third-generation Machine Learning algorithm based on a growing network of sparsely connected Artificial Spiking Neurons. Key points of this model are his growing topology, his interpretability, and the integration of spiking neurons. Models generated with this algorithm can hierarchically represent temporal data and can be merged to create a super-model called a Neural Cloud.