Publicação
Towards the Conceptualization of Refinement Typed Genetic Programming
| Resumo: | The Genetic Programming (GP) approaches typically have difficulties dealing with the large search space as the number of language components grows. The increasing number of components leads to amore extensive search space and lengthens the time required to find a fitting solution. Strongly Typed Genetic Programming (STGP) tries to reduce the search space using the programming language type system, only allowing typesafe programs to be generated. Grammar Guided Genetic Programming (GGGP) allows the user to specify the program’s structure through grammar, reducing the number of combinations between the language components. However, the STGP restriction of the search space is still not capable of holding the increasing number of synthesis components, and the GGGP approach is arguably usable since it requires the user to create not only a parser and interpreter for the generated expressions from the grammar, but also all the functions existing in the grammar. This work proposes Refinement Typed Genetic Programming (RTGP), a hybrid approach between STGP and RTGP, which uses refinement types to reduce the search space while maintaining the language usability properties. This work introduces the ÆON programming language, which allows the partial or total synthesis of refinement typed programs using genetic programming. The potential of RTGP is presented with the usability arguments on two use cases against GGGP and the creation of a prototype propertybased verification tool, pyCheck, proof of RTGPs components versatility. |
|---|---|
| Autores principais: | Santos, Paulo Alexandre Canelas dos |
| Assunto: | Programação Genética Tipos Refinados Síntese de Programas Programação Genética com Tipos Refinados Teses de mestrado - 2020 |
| Ano: | 2020 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | The Genetic Programming (GP) approaches typically have difficulties dealing with the large search space as the number of language components grows. The increasing number of components leads to amore extensive search space and lengthens the time required to find a fitting solution. Strongly Typed Genetic Programming (STGP) tries to reduce the search space using the programming language type system, only allowing typesafe programs to be generated. Grammar Guided Genetic Programming (GGGP) allows the user to specify the program’s structure through grammar, reducing the number of combinations between the language components. However, the STGP restriction of the search space is still not capable of holding the increasing number of synthesis components, and the GGGP approach is arguably usable since it requires the user to create not only a parser and interpreter for the generated expressions from the grammar, but also all the functions existing in the grammar. This work proposes Refinement Typed Genetic Programming (RTGP), a hybrid approach between STGP and RTGP, which uses refinement types to reduce the search space while maintaining the language usability properties. This work introduces the ÆON programming language, which allows the partial or total synthesis of refinement typed programs using genetic programming. The potential of RTGP is presented with the usability arguments on two use cases against GGGP and the creation of a prototype propertybased verification tool, pyCheck, proof of RTGPs components versatility. |
|---|