doktorska disertacija
Abstract
Eksperiment na področju evolucijskega računanja lahko povzamemo s štirimi
pomembnimi koraki: načrtovanje eksperimenta, zagon eksperimenta, analiza rezultatov ter interpretacija rezultatov in diskusija. Vsak korak zahteva posebno pozornost in vsebuje določene pasti, na katere moramo kot raziskovalci biti pozorni.
Disertacija podrobno opiše vse štiri korake, s posebnim poudarkom na statistični
analizi rezultatov, in predstavi novo metodo za primerjavo evolucijskih algoritmov - Chess Rating System for Evolutionary Algorithms (CRS4EAs). Predlagana metoda temelji na šahovskem rangiranju, kjer je vsak evolucijski algoritem predstavljen kot šahovski igralec, vsaka primerjava rešitev dveh algoritmov predstavlja igro med dvema igralcema (in se lahko konča z zmago enega in porazom drugega ali remijem), vsaka parna primerjava med več algoritmi pa predstavlja turnir. Osnova za predlagano metodo je šahovski sistem rangiranja Glicko-2, za katerega tekom disertacije tudi pokažemo, da je najprimernejši. Predlagano metodo skozi velik nabor eksperimentov primerjamo s statističnimi testi z ničelno hipotezo in pokažemo, da lahko s predlagano metodo učinkovito primerjamo uspešnosti evolucijskih algoritmov.
Predlagana metoda najde podobne signifikantne razlike kot bi jih našli z uporabo standardnih statističnih metod, hkrati pa omogoča absolutno vrednotenje moči in uspešnosti algoritmov, ki so vključeni v sistem. Predlagano metodo na učinkovit način uporabimo za uglaševanje parametrov evolucijskega algoritma in jo skozi nabor več eksperimentov primerjamo z drugimi metodami uglaševanja (F-Race in Revac).
Keywords
evolucijski algoritmi;računski eksperiment;ničelna hipoteza;glicko;šahovski rating;
Data
Language: |
Slovenian |
Year of publishing: |
2016 |
Typology: |
2.08 - Doctoral Dissertation |
Organization: |
UM FERI - Faculty of Electrical Engineering and Computer Science |
Publisher: |
[N. Veček] |
UDC: |
004.832:004.8.021(043.3) |
COBISS: |
286228224
|
Views: |
1323 |
Downloads: |
172 |
Average score: |
0 (0 votes) |
Metadata: |
|
Other data
Secondary language: |
English |
Secondary title: |
Chess rating system for evolutionary algorithms |
Secondary abstract: |
Experiment in the field of evolutionary computing can be summarized with four
important stages: experimental design, experiment, analysis, and interpretation
of the results and discussion. Each of these stages requires careful attention to
avoid the pitfalls. The thesis describes all four stages with special emphasis on
the statistical analysis of the results and proposes a new method for comparison
and ranking of evolutionary algorithms - Chess Rating System for Evolutionary
Algorithms (CRS4EAs). The proposed method is based on chess ranking where
each evolutionary algorithm is treated as a chess player, a comparison of the results
of two algorithms is treated as one game between these two algorithms (with three
possible outcomes: win, loss, or draw), and pairwise comparison of all algorithms
is treated as a tournament. The basis of the proposed method is Glicko-2 chess
rating system, which showed as the most appropriate one. We have conducted
a large number of experiments through the thesis in which the proposed method
was compared to different statistical tests of significance. We have shown that
the proposed method is appropriate for comparison and ranking of evolutionary
algorithms. The proposed method finds similar significant differences as statistical
tests, but it also measures the absolute power of algorithms participating in the
system. The proposed method can be used as a method for tuning the parameters
of an evolutionary algorithm, which was also shown through experiments in which
the proposed method was compared to other tuning methods (F-Race and Revac). |
Secondary keywords: |
evolutionary algorithms;computational experiment;null hypothesis;glicko;chess rating;Evolucijski algoritmi;Disertacije; |
URN: |
URN:SI:UM: |
Type (COBISS): |
Dissertation |
Thesis comment: |
Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko |
Pages: |
IX, 213 str. |
ID: |
9169757 |