Titulo Estágio
Strongly-typed Geometric Syntactic Genetic Programming
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
DEI-FCTUC
Enquadramento
Genetic Programming (GP) [1] is a family of Evolutionary Algorithms, a type of Artificial Intelligence methods that are loosely inspired by the principles of evolution by Natural Selection and Mendel’s Genetics. GPs have been successfully applied to difficult problems, from various domains. In particular, we are interested in their ability to automatically generate efficient computer programs that can be executed to solve a given task. Geometric Syntactic Genetic Programming (GSynGP) [2] is a recently proposed GP algorithm that performs geometric crossover operations between the individuals, implicitly controlling the growth of the solutions. It was successfully applied to evolving controllers for robots, tasked with locating the sources of pollutants [3]. The resulting controllers are compact, making them highly interpretable by humans, which not only aids in assessing their soundness as well as enables drawing knowledge from them.
1 - Eiben, A. E., & Smith, J. E. (2015). Introduction to evolutionary computing. Springer-Verlag Berlin Heidelberg.
2 - Macedo, J., Fonseca, C. M., & Costa, E. (2018, March). Geometric crossover in syntactic space. In European Conference on Genetic Programming (pp. 237-252). Cham: Springer International Publishing.
3 - Macedo, J., Marques, L., & Costa, E. (2020). Locating odour sources with geometric syntactic genetic programming. In Applications of Evolutionary Computation: 23rd European Conference, EvoApplications 2020, Held as Part of EvoStar 2020, Seville, Spain, April 15–17, 2020, Proceedings 23 (pp. 212-227). Springer International Publishing.
Objetivo
The main goal of this dissertation is improving the generality of GSynGP. We intend to do it in two ways:
- First, enable GSynGP to evolve programs using functions with a varying number of arguments (i.e., various arities).
- Second, improve the efficiency of GSynGP by introducing types (i.e., strongly-typed GSynGP).
In order to successfully improve GSynGP, the existing rules need to be improved to ensure the validity of the resulting individuals. This not only applies to its existing version, but the problem becomes more difficult with the inclusion of various arities and types.
The resulting approach shall be compared with the existing ones in commonly used benchmark problems.
Plano de Trabalhos - Semestre 1
- Literature review.
- Adaptation of GSynGP to cope with symbols (functions) of various arities (number of arguments) and its preliminary validation.
- Writing of the intermediate report.
Plano de Trabalhos - Semestre 2
- Development of strongly-typed GSynGP.
- Validation of the developed methods.
- Writing of the dissertation.
- Writing of a scientific article with the main results.
Condições
The work shall be carried out within CISUC’s Bio-Inspired Artificial Intelligence group (bAI), under the supervision of Prof. João Macedo and Prof. Ernesto Costa. Additionally, eligible students may have the opportunity to receive a scholarship (Bolsa de Investigação para Licenciado) following the monthly stipend guidelines set by Fundação para a Ciência e Tecnologia (FCT).
Orientador
João Macedo / Ernesto Costa
jmacedo@dei.uc.pt 📩