No abstract

The article presents a study on a new type of an oscillatory neural network (ONN) that uses multilevel neurons for pattern recognition on the basis of high order synchronization effect. The feature of this network architecture is a single oscillator (neuron) at the output with multilevel variation of its synchronization value with the main oscillator thus allowing classifications of an input pattern into a set of classes. ONN model is realized on thermally-coupled VO2-oscillators. ONN training was performed by using the trial-and-error method for the network parameters selection. It is shown that ONN is capable to classify 512 visual patterns (as a cell array 3x3, distributed by symmetry into 102 classes) into a set of classes with the maximum number of elements up to P=11. Classification capability of a network is studied depending on the interior noise level and synchronization effectiveness parameter. The obtained results allow designing multilevel output cascades of neural networks with high net data throughput and it may be applied in ONNs with various coupling mechanisms and oscillators topology.

No abstract

No abstract