The Formation of Computer Cluster with Limited Computing Resources Based on an Arbitrary Neural Network

We studied and compared the application areas of computer clusters with limited computing resources and well-known single-board computers. Existing solutions for calculating convolutional neural networks are analyzed. The possibility of implementing convolutional neural networks on computer clusters with limited computing resources has been established. The system implements an architecture that allows you to organize and distribute convolutional neural network calculations without dividing the convolution sphere into parts with load balancing taking into account the specifics of the data and resources processed on the cluster nodes. The developed algorithm for the application of convolutional neural networks by a computer cluster with limited computing resources makes it possible to use the developed forecasting models of convolutional neural networks to perform calculations in cyberphysical systems of the Internet of Things.