FRANKEN: Automated Generation of Optimization Meta-Heuristics
Did you know that there are over 200 different meta-heuristic optimization algorithms? I have collected some of the crazier one in the Evolutionary Computation Bestiary: Bat algorithm, Elephant Algorithm, Wolf Algorithm, Black Hole algorithm...
The truth is that most of those algorithms are not really that different. Surely some have a cool idea here or there (I really like cuckoo search), but a large majority of them are just small variations of GA or ES.
But if that's the case, could we create a system that break down those algorithms for their parts, and remix them automatically to make something even better? Enter the FRANKEN project:
The franken project is broken up in two parts: 1- How can we systematically break down meta-heuristic search algorithms into a library of common parts? 2- How can we reassemble these parts into new, effective algorithms?
I expect that by studying these two problems, we can gain important insights into the nature of meta-heuristic search, such as:
- What are the important parts of meta-heuristic search algorithms?
- Are the several existing algorithms different in a fundamental way? What kind of changes really impact algorithm design?
- Are there areas of the design space of algorithms that haven't been explored by human designers yet?
But, in the end, the main draw of this project is the joy of breaking down our toys and them putting them back together!
Papers
-
Felipe Campelo, Lucas Batista, Claus Aranha, "The MOEADr Package - A Component-Based Framework for Multiobjective Evolutionary Algorithms Based on Decomposition", Journal of Statistical Software, Volume 92, Issue 6, 2020.2, DOI: 10.18637/jss.v092.i06
-
Jair Pereira Junior, Claus Aranha, and Tetsuya Sakurai, "A Training Difficulty Schedule for Effective Search of Meta-Heuristic Design", IEEE Congress on Evolutionary Computation, 2020.07 (accepted)
-
Anna Bogdanova, Jair Pereira Junior, Claus Aranha, "Franken-Swarm: Grammatical Evolution for the Automatic Generation of Swarm-like Meta-heuristics", In Genetic and Evolutionary Computation Conference Companion (GECCO '19 Companion), DOI: 10.1145/3319619.3321902, 2019.7
-
Claus Aranha, Jair Pereira Junior, Hitoshi Kanoh, "Comparative Study on Discrete SI Approaches to the Graph Coloring Problem", Proceedings of the Annual Genetic and Evolutionary Computation Conference, Proceedings Companion pp.81-82, 2018.7, Conference Proceedings
Other information
I have tried to submit this project for a kakenhi (japanese NSF) grant 5 times, and got rejected every time. Help! :-(