ARTIFICIAL NEURAL NETWORK DEVELOPMENT USING P2P NETWORKS

Abstract

There are many choices when creating an effective artificial neural network for a given problem. The artificial neurons and connections to be used, the network topology, the learning algorithm, the problem representation, all with unique parameters. These choices are continuously enhanced by neural network researchers. The approach taken in this paper is to search through this space of choices and associated parameters using human guidance and experience until an effective combination is found. A peer-to-peer software tool has been implemented to perform this search, called the ANN Grid Toolset. Our toolset allows researchers to share their computing resources, share their learning algorithms and to collectively view the progress of neural network creation. Peer-to-Peer technologies ensures that our toolset is scalable, allowing a large number of resources to be harnessed. We believe that our approach successfully addresses the problem of making appropriate choices for creating an effective artificial neural network.