Figure 6
The difference between the predictions and the true error values, where the underlying function is f(x) = . Each point is obtained by averaging over 40 simulations (generating synthetic data and training the model 40 times). In (a) and (b), the triangles with error-bars are the actual errors of the convolution (a) and the underlying function (b), and the diamonds with error-bars represent the uncertainty inferred by the model using equation (19). In (c), (e), (d) and (f), diamonds, circles, squares and triangles are used respectively as the varied parameter increases in size. In (c) and (d), the measurement error () is inferred from the data, whereas in (e) and (f) it is input as prior knowledge. |