From Explanation to Design: Principled Generators in Program Trace Optimization

Created by W.Langdon from gp-bibliography.bib Revision:1.9181

@InProceedings{Moraglio:2026:ECXAI,

• author = "Alberto Moraglio",
• title = "From Explanation to Design: Principled Generators in Program Trace Optimization",
• booktitle = "Evolutionary Computing and Explainable Artificial Intelligence (ECXAI)",
• year = "2026",
• editor = "Jaume Bacardit and Alexander Brownlee and Stefano Cagnoni and Martin Fyvie and Giovanni Iacca and David Walker",
• address = "Trento, Italy",
• month = "29 " # aug,
• keywords = "genetic algorithms, genetic programming, Program Trace Optimization, PTO, Genotype-Phenotype Map, Locality and Modularity, Principled Design, Explainable Evolutionary Computation, Causal Analysis, logarithmic mean pleiotropy under mutation, genotype-phenotype map",
• URL = "https://ecxai.github.io/ecxai/ppsn2026/PPSN-ECXAI_2026_paper_1.pdf",
• URL = "https://ore.exeter.ac.uk/ndownloader/files/56766944",
• size = "14 pages",
• abstract = "Program Trace Optimization (PTO) is a general framework in which users write probabilistic generators in a problem's natural representation, and generic solvers operate on the generators execution traces as a universal genotype. PTO has long claimed, informally, that writing generators in a well-structured way induces meaningful phenotypic evolutionary operators, but the claim has lacked theoretical grounding. A recent analytical theory has provided a causal account of PTOs genotype-phenotype map that, given a generator, predicts the quality of the induced operators. This paper uses the same theory generatively. Starting from the desired operator properties of locality and modularity, we derive concrete code-level conditions on the generator that induce them. These conditions align with principles long established in pure functional programming, structured programming, and algebraic data types. The broader point is methodological: a general explanatory theory can be used predictively or generatively, and the generative mode yields principled design guidance.",
• notes = "https://ecxai.github.io/ecxai/workshop-ppsn-2026.html Workshop on Evolutionary Computing and Explainable AI @PPSN 2026",

}

Genetic Programming entries for Alberto Moraglio

Citations