Genetic programming applied to RFI mitigation in radio astronomy
Created by W.Langdon from
gp-bibliography.bib Revision:1.8098
- @MastersThesis{Staats:2016:mastersthesis,
-
author = "Kai Staats",
-
keywords = "genetic algorithms, genetic programming",
-
title = "Genetic programming applied to RFI mitigation in radio
astronomy",
-
school = "University of Cape town",
-
year = "2016",
-
type = "Master of Science",
-
address = "South Africa",
-
month = dec,
-
keywords = "genetic algorithms, genetic programming",
-
URL = "https://open.uct.ac.za/handle/11427/23703",
-
URL = "https://open.uct.ac.za/bitstream/item/26627/thesis_sci_2016_staats_kai.pdf",
-
size = "153 pages",
-
abstract = "Genetic Programming is a type of machine learning that
employs a stochastic search of a solutions space,
genetic operators, a fitness function, and multiple
generations of evolved programs to resolve a
user-defined task, such as the classification of data.
At the time of this research, the application of
machine learning to radio astronomy was relatively new,
with a limited number of publications on the subject.
Genetic Programming had never been applied, and as
such, was a novel approach to this challenging arena.
Foundational to this body of research, the application
Karoo GP was developed in the programming language
Python following the fundamentals of tree-based Genetic
Programming described in A Field Guide to Genetic
Programming by Poli, et al. \cite{poli08:fieldguide}.
Karoo GP was tasked with the classification of data
points as signal or radio frequency interference (RFI)
generated by instruments and machinery which makes
challenging astronomers ability to discern the desired
targets. The training data was derived from the output
of an observation run of the KAT-7 radio telescope
array built by the South African Square Kilometre Array
(SKA-SA). Karoo GP, kNN, and SVM were comparatively
employed, the outcome of which provided noteworthy
correlations between input parameters, the complexity
of the evolved hypotheses, and performance of raw data
versus engineered features. This dissertation includes
description of novel approaches to GP, such as upper
and lower limits to the size of syntax trees, an
auto-scaling multiclass classifier, and a Numpy array
element manager. In addition to the research conducted
at the SKA-SA, it is described how Karoo GP was applied
to fine-tuning parameters of a weather prediction model
at the South African Astronomical Observatory (SAAO),
to glitch classification at the Laser Interferometer
Gravitational-wave Observatory (LIGO), and to
astro-particle physics at The Ohio State University.",
-
notes = "Supervised by Prof. Bruce Bassett,",
- }
Genetic Programming entries for
Kai Staats
Citations