Combining Quantitative Data with Logic-Based Specifications for Parameter Inference.

Continuous time Markov chains are a common mathematical model for a range of natural and computer systems. An important part of constructing such models is fitting the model parameters based on some observed data or prior domain knowledge. In this paper we consider the problem of fitting model parameters with respect to a mix of quantitative data and data formulated as temporal logic formulae. Our approach works by defining a set of conditions that capture the dynamics inferred by the quantitative data. This allows for a straightforward way to combine the information from the quantitative and logically specified knowledge into one parameter inference problem via rejection sampling.