Titulo Estágio
Automating Machine Learning: A Graph-based Approach
Áreas de especialidade
Sistemas Inteligentes
Local do Estágio
CISUC
Enquadramento
he Evolutionary Computation area appeared in the 1960s, as an Artificial Intelligence branch that draws its inspiration in the principles of natural selection and genetics. Historically, there are several Evolutionary Algorithm (EA) models, such as Genetic Algorithms, Evolutionary Strategies, Evolutionary Programming, or Genetic Programming.
To tackle a certain problem, a set of potential solutions is created and evaluated. Then, variation operators are iteratively applied to the best solutions leading to the appearance of new promising ways to solve the problem at hand. The process stops when a predetermined termination criterion is met.
Genetic Programming (GP) is an EA branch where solutions codify an algorithmic strategy, i.e., a computer program. Algorithms can be encoded using different types of representations, such as parse-trees or grammars.
Cartesian Genetic Programming (CGP) [1] is a form of GP, where the programs are represented using directed acyclic graphs.
Most of the current implementations of CGP rely only on mutation as variation operator, disregarding the use of the crossover operator.
[1] Miller, J. F., & Thomson, P. (2000, April). Cartesian genetic programming. In European Conference on Genetic Programming (pp. 121-132). Springer, Berlin, Heidelberg.
[2] Machado, P., Nunes, H., & Romero, J. (2010, April). Graph-based evolution of visual languages. In European Conference on the Applications of Evolutionary Computation (pp. 271-280). Springer, Berlin, Heidelberg.
Objetivo
- Study and understand CGP.
- Analyse the graph-based crossover detailed in [2].
- Implement the crossover in CGP.
- Experiment and validate the proposed approach.
Plano de Trabalhos - Semestre 1
T1 - Analyse the State of the Art.
T2 - Select of a CGP framework to be used as code base for the project.
T3 - Design of the solution and of the validation strategies.
T4 - Apply CGP to evolve ANNs (without crossover).
T5 - Write the Dissertation Plan.
Plano de Trabalhos - Semestre 2
T1 - Implement the graph-based crossover operator (described in [2]).
T2 - Test and debug of the implementation of the crossover operator.
T3 - Experiment and validate, according to the previously defined strategies.
T4 - Write a scientific publication.
T5 - Write the Dissertation.
Condições
T1 - Implement the graph-based crossover operator (described in [2]).
T2 - Test and debug of the implementation of the crossover operator.
T3 - Experiment and validate, according to the previously defined strategies.
T4 - Write a scientific publication.
T5 - Write the Dissertation.
Observações
The work is to be conducted at the CMS and ECOS laboratories of CISUC. A workplace will be provided as well as the required computational resources.
Orientador
Nuno Lourenço / Filipe Assunção / Penousal Machado
naml@dei.uc.pt 📩