PRACTICAL ASPECTS AND EXPERIENCES - PARALLEL GENETIC PROGRAMMING AND ITS APPLICATION TO TRADING MODEL INDUCTION

This paper presents a scalable parallel implementation of genetic prog ramming on distributed memory machines. The system runs multiple maste r-slave instances each mapped on all the allocated nodes and multithre ading is used to overlap message latencies with useful computation. Lo ad balancing is achieved using a dynamic scheduling algorithm and comp arison with a static algorithm is reported. To alleviate premature con vergence, asynchronous migration of individuals is performed among pro cesses. We show that nearly linear speedups can be obtained for proble ms of large enough size. The system has been applied to infer robust t rading strategies which is a compute-intensive financial application.