Parameter tuning and control in Evolutionary Algorithms - Cenaero - Belgique
Cenaero
Région
Belgique
The Minamo team is dedicated to the development and implementation of the Cenaero's in-house multi-disciplinary optimization platform, named Minamo. Although computing power has increased dramatically in the last past years, computational time is still an issue as more and more complex simulation analysis are required in industrial design processes. Aiming to tackle this numerical challenge, Minamo provides efficient online surrogate-based optimization methods, based on evolutionary algorithms, allowing to quickly gain insight into the conception space, to quantitatively identify key factors and trades and to automatically explore the conception space to find innovative designs options. Indeed, an adequate and general answer to design optimization based on long running and computationally intensive analysis lies in the exploitation of surrogate models in lieu of the expensive analysis results. Minamo implements several variants of mono- and multi-objective evolutionary, efficiently coupled in an online framework (i.e. with continuous enrichment of the construction support along the design iterates) to the surrogate models.
Context
The context of the proposed research work is to improve the convergence rate of the Evolutionary Algorithm (EA) implemented in Minamo.
An evolutionary algorithm is a search heuristic that mimics processes of natural evolution, such as inheritance, mutation, selection, and crossover, in order to reach the best possible solution to a given problem. Weak individuals tend to die before reproducing, while the stronger ones live longer and bear many offsprings, who often inherit the qualities that enabled their parents to survive. EAs are becoming more and more widely used in real-world industrial applications.
An EA is made of several genetic operators, which are operators used to maintain genetic diversity, known as mutation and to combine existing solutions into others, via selection and crossover. The issue of setting the values for the different EA parameters (for instance, related to these operators) is crucial for good performance of the EA. These parameters values can be set in advance (parameter tuning) or changed during evolution (parameter control). This last approach has the potential of adjusting the algorithm to the problem while solving the problem.
Objective
The idea of this work is to:
Perform a bibliographical review on parameter tuning and control in EA;
Select and implement several strategies for parameter choices;
Validate the implementation on mathematical problems;
Compare the different approaches.
The developments will be performed within the Minamo C++ source codes.
The student should have good programming skills as well as a good mathematical background. Working knowledge of Linux and C++ are valuable assets.
The candidate will join the Minamo development team.
Duration
The length of the internship may vary depending on the availability of the candidate.
Contact
Interested candidates should send a cover letter, quoting reference number of the offer, and a resume to [email protected]
«
Back
Context
The context of the proposed research work is to improve the convergence rate of the Evolutionary Algorithm (EA) implemented in Minamo.
An evolutionary algorithm is a search heuristic that mimics processes of natural evolution, such as inheritance, mutation, selection, and crossover, in order to reach the best possible solution to a given problem. Weak individuals tend to die before reproducing, while the stronger ones live longer and bear many offsprings, who often inherit the qualities that enabled their parents to survive. EAs are becoming more and more widely used in real-world industrial applications.
An EA is made of several genetic operators, which are operators used to maintain genetic diversity, known as mutation and to combine existing solutions into others, via selection and crossover. The issue of setting the values for the different EA parameters (for instance, related to these operators) is crucial for good performance of the EA. These parameters values can be set in advance (parameter tuning) or changed during evolution (parameter control). This last approach has the potential of adjusting the algorithm to the problem while solving the problem.
Objective
The idea of this work is to:
Perform a bibliographical review on parameter tuning and control in EA;
Select and implement several strategies for parameter choices;
Validate the implementation on mathematical problems;
Compare the different approaches.
The developments will be performed within the Minamo C++ source codes.
The student should have good programming skills as well as a good mathematical background. Working knowledge of Linux and C++ are valuable assets.
The candidate will join the Minamo development team.
Duration
The length of the internship may vary depending on the availability of the candidate.
Contact
Interested candidates should send a cover letter, quoting reference number of the offer, and a resume to [email protected]
«
Back