- author = "Zsolt Nemeth and Penn {Faulkner Rainford} and Barry Porter",
- title = "Phenotypic Species Definitions for Genetic Improvement of Source Code",
- booktitle = "ALIFE 2024: Proceedings of the 2024 Artificial Life Conference",
- year = "2024",
- editor = "Andres Faina and Sebastian Risi and Eric Medvet and Kasper Stoy and Bert Chan and Karine Miras and Payam Zahadat and Djordje Grbic and Giorgia Nadizar",
- pages = "530--539",
- address = "Copenhagen",
- month = jul # " 22-26",
- organization = "The International Society for Artificial Life",
- publisher = "MIT Press",
- keywords = "genetic algorithms, genetic programming, genetic improvement, speciation, ecospecies, phenotype, hash, Dana, Shannon-index, Simpson-index",
-
URL = "
https://doi.org/10.1162/isal_a_00795",
-
URL = "
https://direct.mit.edu/isal/proceedings-pdf/isal2024/36/65/2461073/isal_a_00795.pdf",
-
DOI = "
doi:10.1162/isal_a_00795",
-
code_url = "https://doi.org/10.5281/zenodo.11477826",
- size = "10 pages",
- abstract = "Emergent software systems are composed of elementary building blocks, where many of those blocks have variations available which are better or worse in different deployment contexts. Genetic Improvement (GI) for source code has been proposed for creating and curating collections of such blocks, but the combination of new code synthesis with genetic mutation and crossover results in large, complex search spaces. A range of methods to aid such a search have been proposed, with the particular notion of species having appeared in the context of Genetic Algorithms (GAs) to identify individuals with similar genotypes for controlling competition, encouraging the exploration of distant local optima, maintaining diversity and avoiding premature convergence. In this paper we examine a species definition for GI for source code, a domain which has specific features: genotype similarity is largely irrelevant; distance between individuals is undefined; and the fitness landscape is extremely rugged. We propose a phenotypic species definition that captures an algorithms functional phenotypic characteristics, while excluding its nonfunctional phenotypic characteristics (and its particular representation in source code). We introduce our proposal in a GI for a hash table scenario, where species are characterised by divergence in probability distributions.",
-
notes = "'better building block variants' 'new species cause
improvements of 15 percent in performance,' 'GI in
emergent software systems'
'probe captures function call parameter values and return values' used with GI to improve chosen building block. 'probability divergence metrics' => 'meaningful diversity of a population'
species of hash functions defined by Kullback-Leibler divergence and so is application-specific but can be applied to 'broader scope of mapping-related problems, such as schedulers and cache eviction policies'.
fitness = execution speed in 1e-06 seconds. 'species fingerprint'
'correlations between fitness changes and species events, and the mutation and crossover effects which tend to lead to species creation and extinction'
pop=50 (2 elite), 200gens. Insert before, Insert after, Modify, Delete
early 'major improvements in fitness are the result of new species (i.e., new distributions of elements in the hashtable)' later 'it may be that minor implementation details determine further fitness improvements'. However 'considerable changes in performance are almost always brought by the arrival of new species.' 'crossover is generally more successful at creating new species' but...
'right species turnover rate' 'from gen. 160 the diversity is negligible.' 'library of genetic material'
