Bayesian multi-parameter estimation using the mechanical equivalent of logical inference

Abstract

In this work we illustrate how the mathematics of rational thinking is formally equivalent to that of structural mechanics. Concepts from the wold of logic, such as accuracy, uncertainty, Maximum a Posteriori (MAP) and rationality correspond, in the world of mechanics, to stiffness, flexibility, equilibrium and conservativeness. For instance, a linear Gaussian N-parameter estimation problem can be solved through a N-dof linear elastic system, as the analogy goes along these lines: the parameters covariance matrix is the system's flexibility matrix

the Fishers information is the stiffness matrix

the negative log-distribution of the parameters is the elastic potential energy of the system

the Maximum a Posteriori (MAP) is the state of static equilibrium. In principle, based on this analogy, we could reproduce any logical inference problem with a finite element model, and make a judgment by finding its equilibrium state. We will show application of this analogy to a number of civil engineering inference problems, including Bayesian estimation, Bayesian networks and Kalman filter.