Figure 5From: Noise-injected neural networks show promise for use on small-sample expression dataEffects of different amounts of noise for uncorrelated linear model. Classification error is plotted as a function of the amount of noise injection. Each line corresponds to a fixed training sample size, from 10 to 100 for every 10 sample points, the highest line corresponding to 10 sample points and the lowest corresponding to 100 sample points.Back to article page