- author = "Guido F. Smits and Ekaterina Vladislavleva and Mark E. Kotanchek",
- title = "Scalable Symbolic Regression by Continuous Evolution with Very Small Populations",
- booktitle = "Genetic Programming Theory and Practice VIII",
- year = "2010",
- editor = "Rick Riolo and Trent McConaghy and Ekaterina Vladislavleva",
- series = "Genetic and Evolutionary Computation",
- volume = "8",
- address = "Ann Arbor, USA",
- month = "20-22 " # may,
- publisher = "Springer",
- chapter = "9",
- pages = "147--160",
- keywords = "genetic algorithms, genetic programming, symbolic regression, continuous evolution, parallel computing, evolvability",
- isbn13 = "978-1-4419-7746-5",
-
URL = "
http://www.springer.com/computer/ai/book/978-1-4419-7746-5",
-
DOI = "
doi:10.1007/978-1-4419-7747-2_9",
-
abstract = "The future of computing is one of massive parallelism.
To exploit this and generatemaximumperformance itwill
be inevitable thatmore co-design between hardware and
software takes place. Many software algorithms need
rethinking to expose all the possible concurrency,
increase locality and have built-in fault tolerance.
Evolutionary algorithms are naturally parallel and
should as such have an edge in exploiting these
hardware features.
In this paper we try to rethink the way we implement symbolic regression via genetic programming with the aimto obtainmaximumscalability to architectures with a very large number of processors. Working with very small populations might be an important feature to obtain a better locality of the computations. We show that quite reasonable results can be obtained with single chromosome crawlers and a diverse set of mutation-only operators. Next we show that it is possible to introduce a mechanism for constant innovation using very small population sizes. By introducing a computation, with competition for cpu-cycles based on the fitness and the activity of an individual, we can get continuous evolution within the same cpu-budget as the single chromosome crawlers. These results are obtained on a real life industrial dataset with composition data from a distillation tower with 23 potential inputs and 5000 records.",
- notes = "part of \cite{Riolo:2010:GPTP}",
