In this paper, we make a first assessment of the performance of ActoDatA, a novel actor-based software library for distributed data analysis and machine learning in Java that we have recently developed. To do so we have implemented an evolutionary machine learning application based on a distributed island model. The model implementation is compared to an equivalent implementation in ECJ, a popular general-purpose evolutionary computation library that provides support for distributed computing. The testbed used for comparing the two distributed versions has been an application of Sub-machine code Genetic Programming to the design of efficient low-resolution binary image classifiers. The results we have obtained show that the ActoDatA implementation is more efficient than the corresponding ECJ implementation.
Island model in ActoDatA: An actor-based implementation of a classical distributed evolutionary computation paradigm / Petrosino, G.; Bergenti, F.; Lombardo, G.; Mordonini, M.; Poggi, A.; Tomaiuolo, M.; Cagnoni, S.. - (2021), pp. 1801-1808. (Intervento presentato al convegno 2021 Genetic and Evolutionary Computation Conference, GECCO 2021 nel 2021) [10.1145/3449726.3463210].
Island model in ActoDatA: An actor-based implementation of a classical distributed evolutionary computation paradigm
Petrosino G.;Bergenti F.;Lombardo G.;Mordonini M.;Poggi A.;Tomaiuolo M.;Cagnoni S.
2021-01-01
Abstract
In this paper, we make a first assessment of the performance of ActoDatA, a novel actor-based software library for distributed data analysis and machine learning in Java that we have recently developed. To do so we have implemented an evolutionary machine learning application based on a distributed island model. The model implementation is compared to an equivalent implementation in ECJ, a popular general-purpose evolutionary computation library that provides support for distributed computing. The testbed used for comparing the two distributed versions has been an application of Sub-machine code Genetic Programming to the design of efficient low-resolution binary image classifiers. The results we have obtained show that the ActoDatA implementation is more efficient than the corresponding ECJ implementation.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.