Adaptive stochastic resonance for noisy threshold neurons based on mutual information

Noise can improve how a threshold neuron converts bipolar input signals into binary outputs. Such favorable use of noise is the so-called "stochastic resonance" or SR effect at the level of idealized spiking neurons. The paper presents theoretical and simulation evidence that (1) many types of noisy threshold neurons exhibit the SR effect in terms of the mutual information between random input and output sequences and (2) a new statistically robust learning law can find this entropy-optimal noise level. Histograms estimate the relevant probability density functions at each learning iteration. The adaptive entropic SR effect occurred for additive noise processes with both finite and infinite variance (impulsive noise). These findings support the implicit SR conjecture that biological neurons have evolved to exploit their noisy environments