Towards Layer-Wise Optimization of Contextual Neural Networks with Constant Field of Aggregation.

In this paper contextual neural networks with different numbers of connection groups in different layers of neurons are considered. It is verified if not-uniform patterns of numbers of groups can influence classification properties of contextual neural networks. Simulations are done in dedicated H2O machine learning environment enhanced with Generalized Backpropagation algorithm. Experiments are performed for selected UCI machine learning problems and cancer gene expression microarray data of bone marrow acute lymphatic and myeloid leukemia.