Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
409485 | Neurocomputing | 2013 | 10 Pages |
A dynamic phase–space constraint method is proposed to control complex chaotic dynamics in a chaotic neural network (CNN), by limiting refractoriness internal states with a time-varying threshold. The limiting threshold evolves according to a control signal derived from the feedback internal states of the network. Simulation results reveal that the CNN under control exhibits multiphase behavior in the control parameter space. With proper parameter values, the controlled CNN converges to a periodic orbit which includes a stored pattern that has the smallest Hamming distance to its initial state. The properties of the controlled CNN can be used for information processing such as memory retrieval and pattern recognition.
► A dynamic threshold method is proposed to control a chaotic neural network (CNN). ► A threshold evolves according to a control signal derived from internal states. ► The controlled CNN shows a multiphase behavior in the control parameter space. ► The CNN can be controlled to the most similar stored pattern to its initial one. ► The controlled CNN can be used for information processing.