Bayesian inductively learned modules for safety critical systems

Abstract

This work examines the use of Bayesian inductively learned software modules for safety critical systems. Central to the safety critical application is the desire to generate confidence measures associated with predictions. This is achieved in this study by casting the problem in a Bayesian formulation, and is implemented using reversible jump Markov Chain Monte Carlo (RJ-MCMC). We use conventional and novel classification architectures, including logistic discriminants, probabilistic k-nn and radial basis function networks. Results from these methods are illustrated on real life critical systems, including medical trauma data. We report results on the trade-off between model complexity and the width of the posterior predictive probability.