- author = "Davide Farinati",
- title = "Enhancing Geometric Semantic Genetic Programming: a Study on Dynamic Population and Deflate Geometric Mutation",
- school = "Information Management School, Universidade NOVA de Lisboa",
- year = "2025",
- address = "Portugal",
- month = jul,
- keywords = "genetic algorithms, genetic programming, Geometric Semantic Genetic Programming, SLIM-GSGP, IGSM, TIE-DP-GSGP-SEI, Machine learning, Feature selection, Genotype, phenotype, Geometry, Genetic algorithms, Semantics, Artificial intelligence, 0800:Artificial intelligence",
- language = "English",
- isbn13 = "9798273368309",
-
URL = "
https://www.proquest.com/dissertations-theses/enhancing-geometric-semantic-genetic-programming/docview/3303134044/se-2",
- size = "209 pages",
-
abstract = "Genetic Programming (GP) is a branch of evolutionary
computation that evolves programs to solve complex,
non-linear problems without requiring predefined
models. A key application of GP is symbolic regression,
which autonomously discovers mathematical expressions
that best represent data patterns. However, standard GP
operators create a rugged fitness landscape, making
optimization challenging. To address these limitations,
Geometric Semantic Genetic Programming (GSGP) was
introduced, employing Geometric Semantic Operators
(GSOs) that induce a unimodal fitness landscape,
enhancing performance. However, GSGP suffers from
excessive bloat, transforming it into a black-box
model. This dissertation presents several advancements
to mitigate this issue.
First, a dynamic population approach is adapted to GSGP, optimizing fitness evaluations and reducing computational cost. Next, a novel algorithm, Semantic Learning algorithm based on Inflate and deflate Mutation (SLIM), is introduced, using a new Deflate Geometric Semantic Mutation (DGSM) operator that generates smaller offspring while preserving an unimodal fitness landscape. Experimental results show that SLIM produces significantly smaller models than GSGP, with performance comparable to or better than traditional GP. A Python library implementing SLIM is also implemented. Further studies explore SLIMs characteristics, including the impact of DGSM on overfitting and feature selection. A comprehensive analysis of the Mutation Step (MS) parameter is also conducted. Additionally, new Geometric Semantic Crossovers (GSCs) are employed to enhance SLIM's evolutionary capabilities. Finally, SLIM is applied to genomic data for predicting drug response, showing SLIM advantage in real use case scenarios.
This research advances genetic programming by improving interpretability, reducing computational costs, and expanding applications in data-driven modelling.",
-
notes = "Doctor in Information Management with a specialization
in Data Science
2025.32463771
Copyright - Database copyright ProQuest LLC; ProQuest does not claim copyright in the individual underlying works; Last updated - 2026-09-18
Supervised by: Leonardo Vanneschi",
