Stimulus classification using chimera-like states in a spiking neural network

Abstract

Abstract A complex network of bistable Hodgkin-Huxley (HH) neurons with excitatory coupling can exhibit a partially spiking chimera behavior. We propose to use this chimera-like state for classification of the entering stimulus amplitude in the neural network with coexisting resting and spiking states. Due to different additive noise applied to each neuron in the network, the neurons are nonidentical. Therefore, depending on the amplitude of the external current, a part of the neurons stays in the resting state, while another part oscillates. Keeping fixed the coupling strength between neurons inside the network, we train the neural network on external pulses with two different amplitudes to adjust the coupling strength between the network neurons and two output neurons. We consider two variants of the classifier, in the presence and in the absence of inhibitory coupling between output neurons, and study how the output neurons respond to the external pulses of different amplitudes. The accuracy of the proposed classifier reaches 100% when the output neurons are inhibitory coupled, so that only one of these neurons is activated.