Genetic programming on program traces as an inference engine for probabilistic languages

Abstract

Methods of simulated annealing and genetic programming over probabilistic program traces are developed firstly. These methods combine expressiveness of Turing-complete probabilistic languages, in which arbitrary generative models can be defined, and search effectiveness of meta-heuristic methods. To use these methods, one should only specify a generative model of objects of interest and a fitness function over them without necessity to implement domain-specific genetic operators or mappings from objects to and from bit strings. On the other hand, implemented methods showed better quality than the traditional mh-query on several optimization tasks. Thus, these results can contribute to both fields of genetic programming and probabilistic programming.