Fig 4. Accuracy and error function vs noise intensity for threshold spoiling and learning resonance.

(a) Accuracy vs. noise intensity for various values of threshold spoiling. Here we see 3 curves for various values of threshold spoiling, T = 0, 0.5, 1.1. The blue line 1 shows the case for zero threshold spoiling, here we see perfect accuracy for no noise, as would be expected but a sharp decrease with the addition of noise. The same remains for spoiling T = 0.5, line 2. The purple line 3 shows a threshold spoiling which exhibited some degree of stochastic resonance, not only do we see a peak at which it exceeds the unspoiled accuracy but we also see it far exceeding the performance of the unspoiled threshold throughout the intensity range examined. Essentially the threshold spoiling has provided some degree of robustness to the noise which is an extremely interesting property in itself. The contrary could also be inferred that the noise is providing some degree of robustness to threshold spoiling by a ‘blurring’ of the lines as seen in Fig 1b. This figure clearly demonstrates that there is a strong relationship between threshold spoiling, input noise and accuracy but how they work together can be highly variable. (b) Graph showing learning resonance for spoiled T = 1.1, w ₁ = 0.7 w ₂ = 0.7.