Region Attractivity Analysis for Multi-Modality Data Driven Recurrent Neural Networks

The multi-modalilty artificial intelligence is the hot research topic. However the existing neural networks (NNs) stability results are only related to constant or periodic inputs. The multi-modality artificial intelligence requires that neural networks must be able to cope with modality signals, i.e., including constant, periodic and aperiodic signals at the same time. Hence, this paper presents sufficient conditions for region attractivity of neural networks driven by modality data, i.e., hybrid external inputs. By using the comparison method and construting the Lyapunov fuctions, global attractivity conditions are given for NNs driven by multimodality data. For verifying the derived results, two three dimensional NN are simulated.