Stochastic resonance for detection of change in neuronal arrays with threshold

In the paper, the problem of change detection is discussed from a viewpoint of stochastic resonance. Survival in an adversarial environment requires animals to detect sensory changes quickly, as well as accurately. So neurons are challenged to discern ldquorealrdquo change in input as quickly as possible while ignoring noise fluctuations. Mathematically, this is a change-detection problem. It has been established that noise can sometimes help some nonlinearities to enhance signal transmission. Can change detection benefit from noise? A classic change detection problem is introduced. We used neuronal arrays with the threshold-like elements to design a suboptimal detector. The result demonstrates that the detector can perform better than linear detector in non-Gaussian noise and has a more simple architecture than the optimal detector. Fewer samples are needed when change is detected, and input changes can be detected more reliably as well as quickly by adding optimum amount noise. Accordingly, the findings support that neuron population have a reliable capability of exploiting ambient noise.