An Agent-Based Modeling Approach for the Exploration of Self-Organizing Neural Networks

In this thesis we present the ABNNSim toolkit for the simulation of biologically inspired neural networks. This work applies the Agent Based Modeling paradigm to the simulation of biological neural networks, allowing rapid development of models, easy addition of features and a richness of expression that is not available with other tools. The focus of the ABNNSim toolkit is modeling neural networks at the network level. This view leads to some loss of fidelity over other computational neuroscience tools, but allows larger, system-level phenomena to be explored. This is motivated by discoveries in the area of complex networks, in their analysis and prevalence in biological systems. These networks have characteristics that are desirable for biological systems: They are resilient in the face of failure, efficient (both in terms of communication cost and speed of propagation) and can be constructed using simple local rules. This thesis describes several methods of developing complex network topologies in neural networks using pruning. Throughout the project the Agent-Based Modeling paradigm has been valuable as a vital tool for speeding the development and deployment of simulations of this type.