Naslov
SMOTE Inspired Extension for Differential Evolution

Autori
Bajer, Dražen ; Zorić, Bruno ; Dudjak, Mario

Vrsta, podvrsta i kategorija rada
Radovi u zbornicima skupova, cjeloviti rad (in extenso), znanstveni

Izvornik
Bioinspired Optimization Methods and Their Applications. BIOMA 2022. Lecture Notes in Computer Science / Mernik, M. ; Eftimov, T. ; Črepinšek, M. - : Springer, 2022, 216-231

ISBN
978-3-031-21094-5

Skup
10th International Conference on Bioinspired Optimization Methods and Their Applications (BIOMA 2022)

Mjesto i datum
Maribor, Slovenija, 17.11.2022. - 18.11.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
Auxiliary search mechanism ; Differential evolution ; Oversampling ; SMOTE

Sažetak
Although differential evolution (DE) is a well established optimisation method, proven on a wide variety of problems, modifications are proposed on a regular basis attempting to ever more improve its performance. Typical avenues for improvement include the introduction of new (mutation) operators or parameter control schemes. Another, less common approach, is the incorporation of additional, complementary, search mechanisms. This paper proposes one such mechanism, based on the idea of producing new solutions akin to the manner of the SMOTE algorithm producing synthetic minority instances in supervised machine learning. The conducted experimental analysis showed it to be highly competitive against comparable mechanisms on the CEC2014 benchmark suite when incorporated into standard DE, whilst being especially beneficial on simpler multimodal problems. Its incorporation into improved DE variants, although still undoubtedly bringing value on these problems, does hint at complex interactions with already integrated enhancements, suggesting that extending already enhanced algorithm variants is not simple, to say the least.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo

POVEZANOST RADA