Fig. 3.

Schematic flowchart for training and inference using change models. The blue, white and green blocks indicate user defined inputs, intermediate variables and outputs respectively. In the training phase for each parameter change, samples that are drawn from the provided prior distribution are passed through the forward model to estimate pairs of measurements and derivatives. Then, regression models are trained to estimate the distribution of derivatives given the measurements using a maximum likelihood estimation. This phase does not require real data and needs to be done only once. In the inference stage using these trained models we estimate the distribution of the derivatives for any given baseline measurements. We then calculate the posterior probability that change in each parameter caused the change in the measurements using the derivative distributions. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)